Design and development of comprehensive digital solutions with application in the medical field: Human sleep analysis

UNIVERSIDAD DE SEVILLA DEPARTAMENTO DE LENGUAJES Y SISTEMAS INFORMÁTICOS Design and development of comprehensive digital solutions with application in the medical field: Human sleep analysis International Doctoral Dissertation presented by Maksym Gaiduk advised by Dr. Ralf Seepold Dr. Juan Antonio Ortega Ramírez February 2022 ii iii Acknowledgements I would like to express my gratitude to my PhD supervisors Prof. Dr. Ralf Seepold and Prof. Dr. Juan Antonio Ortega Ramírez. They made this result possible through their shared experience, attention, and time invested in this work. Thanks to their constant support, I have been able to develop myself not only as a researcher but also as a person, which is of great importance in life. A big thank goes to my parents and grandparents who have always believed in me, and thanks to them, I have become the person I am. I am also grateful to my brother for the fact that I can always be sure that I can rely on his help in any situation. I would like to express special thanks to my wife Silvia, for whom the time of my doctorate was not always easy, but who has nevertheless kept me motivated, supported, and understood me the whole way from the beginning. Without her, I would not have been able to accomplish this task, and therefore her role cannot be valued highly enough. Of course, I would also like to thank my son for filling my life with new meaning, and thanks to him, my motivation has increased even more. To all my current and former work colleagues, I would like to thank you for being able to put up with me and giving me numerous pieces of advice. I won't be able to name all of you here, but special thanks go to Daniel, who has been two metres ahead of me in the office all these years but has also always been available outside of it to support me. I would also like to thank Agnes for contributing many helpful ideas, especially in the early years, and always being available to talk to. I would also like to thank Lucas for his support and Patrick for motivating me every time we met. Prof. Dr. Natividad Martínez Madrid also played an important role in my development - first during my study, but also afterwards. Thanks to her, I have met several people who have changed my life, and I thank her for that. The Berlin colleagues from Charité, Prof. Dr. Thomas Penzel, and Dr. Martin Glos - thank you for sharing your experience and knowledge. It was essential for the research work that was done. I would also like to thank my colleagues from Italy – Prof. Dr. Massimo Conti and Prof. Dr. Simone Orcioni, for always helping me research-wise and as friends. Also, in Spain, there was always support from colleagues - special thanks to Juan José, but also from my friends and relatives of my wife, who are now also part of my family. Thank you to everyone I have mentioned but also to everyone else I have met in my life – many of you had some impact on it. iv v ABSTRACT The influence of sleep on human life, including physiological, psychological, and mental aspects, is remarkable. Therefore, it is essential to apply appropriate therapy in the case of sleep disorders. For this, however, the irregularities must first be recognised, preferably conveniently for the person concerned. This dissertation, structured as a composition of research articles, presents the development of mathematically based algorithmic principles for a sleep analysis system. The particular focus is on the classification of sleep stages with a minimal set of physiological parameters. In addition, the aspects of using the sleep analysis system as part of the more complex healthcare systems are explored. Design of hardware for non-obtrusive measurement of relevant physiological parameters and the use of such systems to detect other sleep disorders, such as sleep apnoea, are also referred to. Multinomial logistic regression was selected as the basis for development resulting from the investigations carried out. By following a methodical procedure, the number of physiological parameters necessary for the classification of sleep stages was successively reduced to two: Respiratory and Movement signals. These signals might be measured in a contactless way. A prototype implementation of the developed algorithms was performed to validate the proposed method, and the evaluation of 19324 sleep epochs was carried out. The results, with the achieved accuracy of 73% in the classification of Wake/NREM/REM stages and Cohen's kappa of 0.44, outperform the state of the art and demonstrate the appropriateness of the selected approach. In the future, this method could enable convenient, cost-effective, and accurate sleep analysis, leading to the detection of sleep disorders at an early stage so that therapy can be initiated as soon as possible, thus improving the general population's health status and quality of life. vi vii TABLE OF CONTENTS I PREFACE .............................................................................................................. 1 1 INTRODUCTION .................................................................................................. 3 1.1 Research Motivation ....................................................................................... 4 1.2 Research Methodology .................................................................................... 7 1.3 Research Question ........................................................................................... 8 1.4 Research Objectives ........................................................................................ 8 1.5 Success Criteria ............................................................................................... 8 1.6 Thesis Outline ................................................................................................. 9 2 SLEEP ANALYSIS SYSTEMS ........................................................................... 11 2.1 Introduction ................................................................................................... 12 2.1.1 Sleep Disorders ........................................................................................ 12 2.1.2 Arts of Sleep Parameters’ Measurement ................................................. 12 2.1.3 Physiological Signals ............................................................................... 13 2.1.4 Sleep Stages ............................................................................................. 14 2.2 Hardware architecture ................................................................................... 15 2.2.1 Classical Approaches ............................................................................... 16 2.2.2 Alternative Techniques ............................................................................ 17 TABLE OF CONTENTS ___________________________________________________________________________ viii 2.3 Algorithmic Approaches in Sleep Analysis .................................................. 20 2.3.1 Sleep apnoea identification ...................................................................... 20 2.3.2 Sleep stages classification ....................................................................... 21 II SELECTED RESEARCH WORKS ................................................................ 25 3 Automatic sleep stages classification using respiratory, heart rate, and movement signals .......................................................................................................................... 27 4 Digital Health and Care Study on Elderly Monitoring ......................................... 45 5 Estimation of Sleep Stages Analyzing Respiratory and Movement Signals ........ 67 III OTHER RELEVANT RESEARCH WORKS ............................................... 79 6 Recognition of Sleep/Wake States analyzing Heart Rate, Breathing and Movement Signals 81 7 Comparison of sleep characteristics measurements: a case study with a population aged 65 and above 89 8 Home Hospital e-Health Centers for Barrier-Free and Cross-Border Telemedicine 101 9 Recognizing Breathing Rate and Movement While Sleeping in Home Environment 115 10 Embedded system for non-obtrusive sleep apnea detection 125 IV FINAL REMARKS ........................................................................................ 133 11 CONCLUSIONS AND FUTURE WORK ...................................................... 135 11.1 Conclusions ................................................................................................. 136 11.2 Future Work ................................................................................................ 137 A CURRICULUM ............................................................................................. 139 A.1 Research Papers........................................................................................... 139 A.1.1 JCR indexed Journals ........................................................................... 139 A.1.2 Other scientific articles (incl. conference proceedings) ....................... 140 A.2 Awards ........................................................................................................ 144 A.3 Research exchanges..................................................................................... 144 ___________________________________________________________________________ ix A.4 Projects participation ................................................................................... 145 A.5 Open Science ............................................................................................... 146 A.6 Others .......................................................................................................... 147 BIBLIOGRAPHY .................................................................................................... 149 CHAPTER 1: INTRODUCTION ___________________________________________________________________________ 4 1.1 Research Motivation Each of us spends a significant part of our lives in sleep. Nevertheless, the importance of sleep for good health is still often underestimated, although numerous studies have already demonstrated that the influence of sleep on a well-functioning including mental, psychological, and physiological state of the human body is essential [1, 2]. Therefore, it is of great importance to study and to analyse sleep and in case of abnormalities to initiate appropriate treatment. Various sleep disorders can lead to health problems if left untreated. One of them is, for example, chronic insomnia, which affects between 10% and 30% of adults all over the world [3]. In the case of the elderly, the observed incidence is even higher - according to [4], between 30% and 48% are suffering from it [4]. Another common sleep disorder is obstructive sleep apnoea (OSA) which is one of the sleep-related breathing disorders. According to current estimates, it affects about 936 million people between the ages of 30 and 69 worldwide using an Apnoea-Hypopnoea-Index (AHI) criterion of five or more events per hour based on the American Association of Sleep Medicine (AASM) criteria [5]. Restless legs syndrome (RLS) is as well a widespread sleep disorder with the general prevalence rates ranging from 4% to 29% of adults, averaging 14.5 ± 8.0% across different available studies [6]. In addition, there are several other sleep disorders that should be identified and treated in the affected population whenever possible to minimise possible negative health consequences [7]. Further information regarding sleep disorders can be consulted in section 2.1. Besides the health impact, sleep disorders also bring significant economic consequences. Several factors are responsible for this, and some are not obvious, though they should nevertheless be considered. As an example, we can consider the data of incurred costs due to inadequate sleep from Australia, presented in Table 1 for the 2016-2017 financial year [8]. The table provides information on various categories of costs and their origins and includes both financial and non-financial ones for the comprehensive representation. ___________________________________________________________________________ 5 Table 1.1: Breakdown of the costs of inadequate sleep in Australia by various categories (the financial year 2016–2017) [8]. Costs of various categories of inadequate sleep including costs of conditions associated with them Total EDS-SD EDS-Other Insufficient Sleep ($ billions) ($ billions) ($ billions) ($ billions) Financial costs ($ billions) Health 0.50 0.52 0.22 1.24 Productivity Reduced employment 1.27 2.69 1.26 5.22 Premature death 0.24 0.26 0.11 0.61 Absenteeism 0.36 0.94 0.43 1.73 Presenteeism 0.73 2.22 1.68 4.63 Subtotal 2.60 6.11 3.48 12.19 Informal care 0.11 0.18 0.12 0.41 Other (nonmedical accident costs) Workplace accidents 0.05 0.16 0.08 0.29 Vehicle accidents 0.36 0.98 0.85 2.19 Subtotal 0.41 1.14 0.93 2.48 Deadweight loss 0.38 0.75 0.43 1.56 Total financial costs 4.00 8.71 5.17 17.88 Nonfinancial costs ($ billions) Loss of well-being 21.41 5.14 0.78 27.33 Total costs ($ billions) Financial + Nonfinancial 25.41 13.85 5.95 45.21 EDS = excessive daytime sleepiness; SD = sleep disorders. For detailed sleep analysis and evaluation of the individual sleep stages, a standard procedure called polysomnography (PSG) is typically used and performed according to the guidelines of the American Academy of Sleep Medicine (AASM) [9]. Several physiological signals are recorded continuously during sleep when performing a PSG procedure [10]. Each of the above signals undergoes changes depending on the current sleep phase. Evaluation of these changes thus allows classification of the individual's stage of sleep at any given time. More details on PSG measurement are presented in section 2.2.1. Despite the feasibility of an accurate analysis of sleep using PSG, some aspects should be considered that bring with them several disadvantages as well as complications: CHAPTER 1: INTRODUCTION ___________________________________________________________________________ 6 • A sleep laboratory where PSG recording is performed is not a natural sleep environment for patients. In addition, several electrodes, sensors, and cables are attached to the subjects' bodies, as can be seen in Figure 1.1, causing discomfort that can affect sleep patterns [11]. • At least 22 analysis connectors are needed for the analysis in a standard case, which can also cause logistical and economic cost problems [12]. • The time and personnel costs required for recording, processing, and analysing the data are high [13]. • The number of available sleep laboratories and sleep physicians is limited [14], resulting in a delay in the care of patients with sleep disorders. However, this care is crucial as a large number of undiagnosed sleep disorders can lead to serious health problems [15]. These issues indicate that PSG remains a limited method and that several aspects require improvement in the development of sleep analysis systems. Figure 1.1: Typical polysomnography measurement1 1 http://www.nhlbi.nih.gov/health/dci/images/sleep_studies.jpg ___________________________________________________________________________ 7 It is evident from several scientific works that physiological parameters, such as breathing or movement, also vary during different sleep phases and could be used for the identification of various sleep disorders [16, 17]. These signals could potentially be recorded with less effort and more convenience for patients. However, such a solution requires the expertise of several disciplines: medicine, signal acquisition, processing, and analysis. Nonetheless, such development remains reasonable and promising, as, in the future, this approach should enable convenient, cost-effective, accessible to everyone and accurate sleep analysis, leading through broad application to the detection of possible sleep disorders at an early stage so that therapy can be initiated as soon as possible, thus improving the health status and quality of life of the general population. 1.2 Research Methodology The methodology of this research is based on the standard scientific research technique [18], was adapted according to the objective, and includes the following phases: 1. Research on state of the art. During the first phase, the literature review was performed in order to specify the research question and to get a comprehensive overview of existing methodologies, systems, and scientific backgrounds in the field of sleep study. 2. Conception/Modelling. Based on the previously performed literature analysis, the first concepts and mathematical models for sleep analysis were designed in this phase. 3. Prototyping. According to theoretically developed/improved models, the prototype implementation of the sleep analysis system was performed. 4. Development and execution of studies. Research studies with the developed prototype were designed to facilitate the following evaluation of the implemented prototype. 5. Evaluation/analyses of results. The obtained results of the prototype work were evaluated in this phase. 6. Improvement of technology/model. According to the evaluation results, improvements were made to the technologies/model to enhance the outcomes and increase the scientific value of the research. CHAPTER 1: INTRODUCTION ___________________________________________________________________________ 8 Phases 3-6 have been repeated to improve the quality of the research outcome. 1.3 Research Question The following research question that leads this doctoral dissertation was determined during the first phase of the used methodology: What mathematically reasoned methodologies have the potential to be evolved and deployed to provide convenient, affordable, and accurate sleep analysis? From this, research objectives can be derived, which are outlined next. 1.4 Research Objectives This doctoral thesis aims to create a solid scientific basis for developing sleep analysis systems for widespread use, thereby improving the well-being and health condition of the population in the long term. These are among others: • Prepare the scientific background for developing an algorithm for the classification of sleep stages based on signals that can be acquired non-invasively. • Determine a minimal set of physiological parameters required for the distinction of sleep stages with a sufficient level of accuracy. It must be considered that the signals should be recorded in the absence of direct contact with the body. • Design the features for further algorithmic processing based on the analysis of physiological signals during sleep and their correlations with the current sleep stages. • Identify which type of systems can be used for a non-obtrusive recording of the relevant signals detected during sleep. • Verify whether the same type of systems can be applied for a broader range of sleep-related analytics, including for instance the detection of sleep apnoea. 1.5 Success Criteria Success is achieved when the research question is addressed, which includes completing the previously defined objectives of the research. For this purpose, scientific studies should have been conducted to evaluate the developed mathematically reasoned methodologies in the area of a sleep study. The evaluation should have shown the applicability of the selected methodology of sleep analysis for achieving qualitatively adequate results. ___________________________________________________________________________ 9 Translated to the exact goals of this doctoral research, the success would be the detection of sleep stages at the level that exceeds state of the art, which can concern both the set of parameters used and the accuracy of detection. 1.6 Thesis Outline The structure of the doctoral thesis is as follows: Chapter 2 presents the problem of existing methodologies and algorithmic approaches in sleep analysis and provides the relevant basic information on the topic of this doctoral dissertation. In Part II are presented the selected published journal articles that directly address the research question and the objectives of the dissertation. These articles were published in Clarivate Journal Citation Reports (JCR) journals. Additional published scientific articles are included in Part III. These articles provide further details of the research conducted and present significant additional findings related to the research objectives. The publications include, among others, proceedings of the conference of the category B (CORE) – KES and book chapters in book series from Q3 in SCImago Journal & Country Rank. Finally, in Part IV, conclusions are drawn, important observations are presented, and potential future work is discussed. 10 11 CHAPTER 2 SLEEP ANALYSIS SYSTEMS The idea behind digital computers may be explained by saying that these machines are intended to carry out any operations which could be done by a human computer. Alan M. Turing CHAPTER 2: SLEEP ANALYSIS SYSTEMS ___________________________________________________________________________ 12 2.1 Introduction As described in the previous chapter, sleep is essential for human health. Therefore, in case of irregularities, appropriate therapies should be prescribed. For this, however, the sleep disorders should first be detected, and for this purpose, different approaches can be used. In order to better understand what kind of systems are suitable to be used for this, one should first gain a general understanding of the signals that are to be measured and the disorders that are to be identified. Following is presented the relevant basic information about the physiological signals, arts of measurement, and sleep disorders that are of importance in developing a system for sleep analysis. 2.1.1. Sleep Disorders Currently, there are over 80 known sleep disorders according to the AASM classification [19]. These disorders can be divided into six main categories: • Insomnias • Sleep-Related Breathing Disorders • Central Disorders of Hypersomnolence • Circadian Rhythm Sleep-Wake Disorders • Parasomnias • Sleep-Related Movement Disorders In addition, there is a seventh category: "Other sleep disorders", which includes all other sleep-related symptoms or events that do not meet the standard definition of a sleep disorder. All of these sleep disorders have specific characteristics that allow them to be distinguished. The standards described in detail in [20] should be applied to diagnose them. As a support for this, technical solutions can be used to measure a number of physiological parameters. Subsequently, with appropriate processing and analysis, the recorded signals can be used for the diagnosis. Sections 2.2 and 2.3 further detail possible technical solutions and appropriate algorithms. 2.1.2. Arts of Sleep Parameters’ Measurement The methods of measuring sleep characteristics may be classified into two major groups - objective and subjective measurement [21]. In the case of objective measurement, the relevant values are recorded with electronic devices and evaluated afterwards. For this ___________________________________________________________________________ 13 purpose, various types of sensors can be used, such as a system that is a combination of a pressure sensor and a triaxial accelerometer [22]. This type of system can be used, for example, to detect heart rate, respiration, sleep phases, movement, the position of the person, sleep apnea, and sleep time, and to calculate other parameters from these data using an appropriate algorithm. Besides, a smartphone's sensors can also be used for the objective measurement of sleep quality [23]. However, using of smartphones brings some disadvantages, such as the need to place the phone near the person in the bedroom or the increased battery consumption during the night. In addition, there are differences in the sensors installed in different smartphones, which may have as a result that the measurement data may differ by diverse devices. As for the subjective measurement, it is based on recording the sleep parameters perceived by the subjects. Questionnaires and/or sleep diaries are used for this purpose [24]. Mobile apps can also play the role of electronic sleep diaries. In [25], a comparison of a classic paper-based sleep diary with an electronic version and the actigraphy measurement was made and presented. The evaluation showed that electronic sleep diaries might have an advantage over paper-based versions. The comparison of the results of sleep quality measurements with objective and subjective methods was made in [26]. As a result, a correlation between both types of measurement was confirmed. Further details on subjective and objective measurement of sleep characteristics and the evaluation of the comparison of these two arts of measurement may be found in the article “Comparison of sleep characteristics measurements: a case study with a population aged 65 and above” presented in Part III of this doctoral dissertation. The performed study has disclosed the correlation between different ways of measurement for several sleep characteristics, indicating where the objective measurement could substitute the subjective one. 2.1.3. Physiological Signals When talking about types of physiological signals relevant for sleep analysis, they can be divided into a couple of main groups. These are the following: • Heart-related signals. • Respiratory signals. • Movement signals (including eye movement). • Brain activity signals. CHAPTER 2: SLEEP ANALYSIS SYSTEMS ___________________________________________________________________________ 20 Figure 2.3: Possible placement of sensors under the person 2.3 Algorithmic Approaches in Sleep Analysis The recorded and pre-processed by the hardware signals are then to be algorithmically processed to be able to draw key conclusions about sleep. This can be, for example, the recognition of sleep phases during sleep or the identification of sleep apnoea events. It is also possible to recognise other sleep-relevant parameters employing a corresponding analysis for subsequent providing of the results to an interested person that might be, for instance, a user directly or a physician. Only sleep apnoea and sleep stages detection methods are presented in the following subsections, as the total number of possible applications with corresponding algorithmic approaches is so large that such description would go beyond the scope. 2.3.1. Sleep apnoea identification Sleep apnoea detection is a prominent field of sleep analysis. As mentioned in section 1.1, OSA is a common sleep disorder that affects hundreds of millions of people [5]. However, central sleep apnoea (CSA) also has a prevalence of 0.9% and can lead to complications [41]. For the detection and differentiation of OSA and CSA (mixed form and other breathing-related events such as hypopnea are also possible), mainly respiratory flow, respiratory effort, and oxygen blood saturation are analysed according to AASM criteria [9]. Snoring or PCO2 (in the case of hypoventilation measurement [9]) may also be of interest, as well as position, cardiovascular information, and detection of sleep state (or at least sleep time) [42]. Traditionally, methods that follow the rules described in [9] are used for detection, which can also be linear algorithms, as the rules are clearly defined. ___________________________________________________________________________ 21 In the case of alternative measurements, where either a smaller number of parameters or other signals (such as tracheal sounds [43] or BCG [44]) are measured, other algorithms may be used that have more complicated background logic. These can be, for example, neural networks [45], Gaussian Processes (GP) [46], or support vector machines (SVM) [47]. Nevertheless, other algorithms as well can be applied. A comprehensive review of 84 original research articles on obstructive sleep apnoea detection is presented in [48], while a general overview of alternative approaches in sleep apnoea diagnosis has been made in [49]. An alternative method for the detection of sleep apnoea events and namely the analysis of respiratory signals recorded without contact using sensors placed under the mattress was also developed within the framework of this PhD thesis and is presented in the article "Embedded system for non-obtrusive sleep apnoea detection" in Part III. 2.3.2. Sleep stages classification The main focus of this PhD thesis was on the detection of sleep phases, which were briefly presented in section 2.1.4. The detailed research conducted on state of the art has shown that there are several groups of methods that allow the signals to be analysed. Furthermore, different signals can also be considered as input for the analysis. Some research describes algorithms that will enable detection of sleep stages based on the signals recorded during traditional polysomnography, described in section 2.2.1. An example of this type of signal is the EEG signal, the analysis of which is described in [50, 51]. Such algorithmic systems allow for relatively high accuracy (86.7% for W, N1, N2, N3, and REM in [50]) but require a recording that cannot be performed in a contactless way which has some disadvantages as described before. Based on the assumption mentioned in section 2.2.2 and confirmed by preliminary prototype development, that the recording should be performed with maximum comfort for the user and thus heart, breath, and motion signals can be acquired, the decision was made to use these three signals as input. When it comes to methods of analysing the signals, for example, an SVM can be used to automatically identify sleep and wake phases, as described in [52, 53]. In [52], wrist and chest actigraphy were analysed in combination with heart rate variability (HRV) for sleep classification. The accuracies obtained for the group of 18 healthy adult subjects are almost the same - 78 % and 77 % for, respectively, chest and wrist actigraphy CHAPTER 2: SLEEP ANALYSIS SYSTEMS ___________________________________________________________________________ 22 in combination with HRV. However, HRV requires very exact identification of R (or J in the case of BCG) peaks, which can be challenging due to the contactless art of measurement favoured for the PhD research. In [53], besides SVM, other classifiers that can be used for sleep phase detection were analysed: K-nearest Neighbor and Naïve Bayes. The best classification results were obtained with the Naïve Bayes classifier with a precision of 70.3%, a recall of 71.1%, and overall accuracy of 72.2%. Neural networks are another group of methods that can be applied to detect sleep stages. Numerous scientific papers have analysed different signals using this approach. The work already mentioned [50] used cascaded long short-term memory recurrent neural networks (LSTM) to identify sleep stages. LSTM is a frequently used method, which is also described in [54] for the analysis of HRV. In general, LSTM can be said to be an advanced version of recurrent neural network (RNN) architecture. One of its advantages is that it can model chronological sequences and their long-term dependencies more precisely than conventional RNNs. Nonetheless, RNNs can also be used to analyse signals to detect sleep phases. An implementation is presented in [55], where heart rate and wrist actigraphy recorded with a wearable were analysed. A review of scientific papers using deep learning techniques for sleep phase detection is presented in [56]. For further analysis of automatic sleep phase detection techniques in the last decades, [57] and [58] can be consulted. Analysing the changes during sleep in the heart, breath, and movement signals that were selected for this PhD research, a regression analysis can be applied to find a correlation with the corresponding sleep phases. Since in this case, there are several input signals that can be presented as independent variables, and as a result, the sleep phases should be detected, which can be presented as a limited number of categories of a dependent variable, one of the variations of logistic regression and namely Multinomial Logistic Regression (MLR) could be selected as one of the appropriate methods [59]. This approach, which brief visualisation is presented in Figure 2.4, was also taken for the development of an algorithm for the detection of sleep phases in this doctoral thesis, among other things because, in this implementation, there is not only a rigid assignment of an epoch to a sleep phase, but in the output, the probabilities of being in each sleep phase are represented for each epoch. This circumstance makes further processing with the improvement of the results easier and more flexible. ___________________________________________________________________________ 23 Figure 2.4: General visualisation of the proposed MLR approach The first results of the research and development of an algorithm for the recognition of sleep and wake states based on the MLR method are presented in Part III of this document in the article "Recognition of Sleep/Wake States analysing Heart Rate, Breathing, and Movement Signals". The detection of wake/light/deep/REM sleep stages was also achieved, as presented in the article "Automatic sleep stages classification using respiratory, heart rate and movement signals" in Part II. A further detailed description of the method used, including the explanation of the calculation of the parameters derived from the input signals to increase the accuracy, can also be consulted in these articles. Considering the circumstance that especially recognition of heart signal can be challenging due to the fact that every heartbeat produces only a small movement, which amplitude is significantly lower than provoked by breathing or body movements signals, the doctoral research has been directed towards finding the possibility of recognising the sleep stages exclusively through the analysis of breathing and movement signals. To do this, it was important to closely examine the variations in physiological signals during sleep and develop appropriate features for the algorithm input. After thorough research, this has been achieved, and the results are presented in Part II in the article "Estimation of Sleep Stages Analyzing Respiratory and Movement Signals". 24 25 PART II SELECTED RESEARCH WORKS 26 27 Automatic sleep stages classification using respiratory, heart rate and movement signals Overview This article deals with the estimation of sleep stages through the analysis of movement, heart, and respiratory signals. For this purpose, parameters/features derived from the signals were selected, calculated, and then used as input for the MLR model. Context This publication resulted from research conducted over the first two years of the PhD study. During these years, thorough literature research was carried out, as well as the first developments of the algorithms were based on the knowledge gained. The experience gathered during the research stays in Seville (Spain), and Ancona (Italy) has significantly contributed to achieving these results. Furthermore, the active work in international projects, such as the IBH Living Lab Active and Assisted Living, has supported the development. Journal information The Journal “Physiological Measurement” was selected for the submission of this paper. This journal is indexed in JCR with the current Impact Factor of 2.833. Detailed ranking for the last five years is presented in the following Table. Category Biophysics Engineering, Biomedical Physiology JCR Year JIF Rank JIF Quartile JIF Rank JIF Quartile JIF Rank JIF Quartile 2020 41/71 Q3 53/89 Q3 41/81 Q3 2019 39/71 Q3 49/87 Q3 43/81 Q3 2018 45/73 Q3 41/80 Q3 48/81 Q3 2017 45/72 Q3 39/78 Q2 55/83 Q3 2016 49/73 Q3 37/77 Q2 50/84 Q3 28 This is the Accepted Manuscript version of an article accepted for publication in Physiological Measurement. IOP Publishing Ltd is not responsible for any errors or omissions in this version of the manuscript or any version derived from it. The Version of Record is available online at https://doi.org/10.1088/1361-6579/aaf5d4. Automatic Sleep Stages Classification Using Respiratory, Heart Rate and Movement Signals Maksym Gaiduk1,3, Thomas Penzel2, Juan Antonio Ortega3, Ralf Seepold1,4 1HTWG Konstanz, Konstanz, Germany 2Interdisciplinary Center for Sleep Medicine of Charité Clinic, Berlin, Germany 3University of Seville, Seville, Spain 4Sechenov University, Department of Information and Internet Technology, Moscow, Russia Abstract Objective: This paper presents an algorithm for non-invasive sleep stage identification using respiratory, heart rate and movement signals. The algorithm is part of a system suitable for long-term monitoring in a home environment, which should support experts analysing sleep. Approach: As there is a strong correlation between bio-vital signals and sleep stages, multinomial logistic regression was chosen for categorical distribution of sleep stages. Several derived parameters of three signals (respiratory, heart rate and movement) are input for the proposed method. Sleep recordings of 5 subjects were used for the training of a machine learning model and 30 overnight recordings collected from 30 individuals with about 27,000 epochs of 30-second intervals each were evaluated. Main results: The achieved rate of accuracy is 72% for Wake, NREM, REM (with Cohen’s kappa value 0.67) and 58% for Wake, Light (N1 and N2), Deep (N3) and REM stages (Cohen’s kappa is 0.50). Our approach has confirmed the potential of this method and disclosed several ways for its improvement. Significance: The results indicate that respiratory, heart rate and movement signals can be used for sleep studies with a reasonable level of accuracy. These inputs can be obtained in a non-invasive way applying it in a home environment. The proposed system introduces a convenient approach for a long-term monitoring system which could support sleep laboratories. The algorithm which was developed allows for an easy adjustment of input parameters that depend on available signals and for this reason could also be used with various hardware systems. Introduction The human body is more fragile than people think. In order to survive, it requires sleep just as much as food, water, or oxygen. This is a basic principle of human physiology that has been borne out by thousands of research studies. In general, sleep is a state where our bodies and minds rest and rejuvenate [SPR09]. It is obligatory for our normal physiological, mental and emotional functioning during the awake hours. The belief that it is possible to have just a couple of hours of sleep a night over a long period of time without suffering any negative consequences is a common misconception [NAT11]. It is categorically beyond doubt that when sleep contains even slight abnormalities, the aftermath can lead to physical illness, psychological problems or an untimely death [LEE12]. A popular misconception is that adults have to sleep at least 7 to 8 hours every night to be rejuvenated properly, while children require far more hours of sleep [NAT11]. However, this is only standard recommended advice; sleep requirements are individual for every person [NAT11]. In addition, getting many hours of sleep does not always guarantee a healthy and rested state, because the crucial point here is not the quantity, but the quality [NAT11]. In order to get reliable data on the quality of a person's sleep, the sleep stages and their sequence and durations have to be analysed. Usually such studies are executed in sleep laboratories. The established standard procedure for sleep stage quantitative evaluation is the overnight polysomnography (PSG) following the guidelines of the American Academy of Sleep Medicine (AASM) [CHO13]. Motivation The execution of PSG generally means that during sleep a human body will be connected to at least 22 laboratory wire attachments [CHO13], which explains why the PSG is quite a costly and time-consuming procedure [ETT10]. Moreover, being in the sleep laboratory is always unfamiliar and disturbing for the subjects, so there could be some discomfort because of the electrodes and the issue of movement limitation. All of these conditions would impact the person’s regular sleep behaviour [LEB01]. It is safe to say that the sleep during the PSG monitoring often has differences compared to sleeping at home. Moreover, the fact, that different sleep stages have an appreciable effect on heart rate, breathing and movement [HAY12, KUR12, LON14], provides good reason to combine these parameters to develop the sleep stage classification algorithm. where x[n] is the input data and M represents the filter length. For the presented implementation the value of parameter M was determined to be 5 according to [CHE03]. Figure 1 shows an overlap of the typically unfiltered (blue) and filtered (red) ECG signals for the same time interval. As the figure indicates, the key pillar of the HPF-method lies in the fact that the filter does not shift the signals on the axis X, which otherwise could cause the occurrence of incorrect time values of the R-peaks. Therefore, the lower frequency noise sources of the ECG signal and the baseline wander will be suppressed with the help of the HPF-method. Figure 1: Overlap of the same ECG signal interval After removing the artefacts from the ECG signal with a HPF-filter, the next stage of calculation is to identify the timestamps of all R-peaks. The following step is to estimate the derived parameters and to store, for further computations, the lists of all derived parameters for every subject in a CSV-File. The values from each list are calculated in ratio to the derived parameter average signal. Each list is sorted in chronological order and every entry contains the values for a 30-second epoch. For instance, in the column HRV are determined values for every epoch, calculated as the difference (in ms) between HRV per actual 30s and the average HRV of the full sleep record. The first 10 minutes of the recording of one subject are presented in Table 3 to illustrate the possible deviation of derived parameters. As can be seen in the table, the deviations can be very diverse even for the same sleep stage, but the results of current research confirm that it is still possible to find some dependencies with appropriate mathematical model. This makes sleep stage recognition based on these parameters a challenging, yet promising method. Table 3: Extract of the file Param.csv, including the results of the first 10 minutes of a subject. Index HB HRV HBI BM TSDM PSDM VBR VIN RA DA AllStages FourStages ThreeStages 1 19,234 19,87 -18,031 361,135 -88,419 -85,571 -22,398 0,974 -26,582 -78,349 1 1 1 2 3,121 3,222 -2,095 -7,022 -3,863 38,145 -27,179 -32,812 -6,312 -17,725 1 1 1 3 -0,101 -0,108 -2,352 -12,769 7,2 0,854 -23,869 -13,121 -11,14 -13,394 1 1 1 4 3,121 3,222 -3,133 -18,379 78,726 -33,214 -21,294 -5,255 19,782 -4,734 1 1 1 5 -0,101 -0,108 -3,37 -29,668 56,493 -28,998 -13,939 14,367 58,445 8,257 2 2 2 6 3,121 3,222 -2,122 -20,911 -35,676 -27,311 -20,559 -19,075 40,053 -0,404 2 2 2 7 -0,101 -0,108 -2,627 -29,736 34,384 -34,394 -15,778 14,78 -27,579 8,257 2 2 2 8 -0,101 -0,108 -0,757 -20,5 -1,307 -11,514 -11,732 13,721 -86,397 -4,734 2 2 2 9 -0,101 -0,108 -1,115 -24,263 -64,183 -63,253 -12,468 -14,441 -88,39 -0,404 1 1 1 10 -3,324 -3,438 2,523 -33,636 -18,537 -52,984 -4,744 25,011 -83,562 12,587 2 2 2 11 -3,324 -3,438 1,897 -37,057 7,611 46,315 5,554 29,109 -8,266 16,918 2 2 2 12 -6,546 -6,768 2,766 -35,552 37,512 -8,966 2,244 31,983 81,59 8,257 3 2 2 13 -0,101 -0,108 1,361 -35,825 26,288 -33,008 -1,434 32,808 18,786 16,918 3 2 2 14 -0,101 -0,108 -2,16 -20,158 42,802 -8,722 -2,17 27,197 -30,529 -4,734 3 2 2 15 25,679 26,53 -20,673 137,134 -84,719 -26,599 -34,535 11,507 -6,427 -61,027 1 1 1 16 -3,324 -3,438 0,36 -23,237 -9,815 -11,252 -15,41 -0,374 -12,098 -4,734 1 1 1 17 -3,324 -3,438 3,234 -31,447 -40,002 -40,316 -12,468 15,027 -50,685 8,257 2 2 2 18 -3,324 -3,438 1,897 -37,673 -50,225 -54,089 -8,79 15,206 -99,004 16,918 2 2 2 19 -3,324 -3,438 1,641 -37,741 -45,882 -47,399 -2,17 18,921 -29,533 16,918 2 2 2 20 -3,324 -3,438 3,519 -34,936 2,249 -23,919 10,335 29,673 -55,628 12,587 2 2 2 The list column ‘Index’ represents the consecutive number of sleep epochs, then values of derived parameters are listed. ‘AllStages’ contains the sleep stages obtained by the experts, according to the RK-method. The column ‘Four-Stages’ shows more results of those specialists; however, the categorization is divided into the stages WAKE (1), Light- and Deep Sleep (2 and 3) and REM (4). The column “ThreeStages” represents the experts’ results which are categorized into the stages WAKE (1), NREM (2) and REM (3). Regression analysis is a complex process and difficult to implement in a research project. Furthermore, algorithms have already been implemented in various programming environments [BOC18], [BR18]. Therefore, there was no need to develop completely new algorithms or implementation strategies. To achieve the goals of this project, free software and the programming language R (v.3.3.1) for statistical computing and graphics were utilized. The use of R was justified because of its ability to provide a wide range of statistical variety (classical statistical tests, linear and nonlinear modelling, time-series analysis, clustering, classification, etc.) and various graphical techniques. Also, R is compatible with and runs on a wide variety of UNIX platforms, Windows and MacOS [RF17]. Integration of R into the developed software was done using the corresponding free access libraries. The merged derived parameter list of five files (representing 5 subject recordings) represents the training data, which then is used as a training set for the MLR model in R. In addition, category variables have to be set as factor variables. ”Wake” was chosen as the reference category for the MLR analysis because it is the only ‘stage’ not included in the bigger category “Sleep” containing REM, Light Sleep and Deep Sleep stages. The multinom method of R software is used to fit the multinomial log-linear models to maximum probability estimations. Finally, the significance of the parameters can be analyzed with the 2-tailed Wald-z test [UCL18]. The most important part of this project is the R prediction-method, called predict. It requires the fitted model and the data frame to predict the probability of each stage belonging to a particular category. The predicted results are supposed to be stored in a separate .xls file. Results This project is focused on analysing the sleep records from the Charité clinic in Berlin2. In the main study, about 230 hours of recordings from the Sleep Medicine Center at Charité were analysed. The data originates from the sleep recordings of 30 individuals (the amount of male and female participants is equal) with no significant health disorders. The average age was 38.5 +/- 14.5 years old and the BMI of participants averaged 24.4 +/- 4.9 kg/m². The sleep stages of each PSG recording were manually measured in 30 second time intervals based on the Rechtschaffen and Kales method. As a result, the classification was made for three stages: WAKE, NREM and REM, and for four stages: WAKE, Light Sleep (i.e., NREM 1 and NREM 2), Deep Sleep (i.e., NREM 3 and NREM 4) and REM. Using real data from a sleep laboratory for system evaluation can help to confirm the accuracy of the theoretically developed algorithm in a practical setting. Therefore, firstly, the results of the statistical estimations with regard to the mean and quartiles of the derived parameters were listed to show their behaviour in the 4 main sleep phases – WAKE (W), Light Sleep (LS), Deep Sleep (DS) and REM (R). In the next step, the MLR model fitting process was described. Finally, the results of the regression analysis in 3 and 4 classification categories were displayed for the provided data. In this project, the recordings were strictly separated into the types of data – training and tests. Firstly, the recordings of five subjects were entered into the model for the Charité-dataset for the training. Then, for classification, the recordings of the remaining 30 subjects were used. Average age, height, weight and state of health of subjects in both datasets had no significant differences. At the same time, the training dataset has also covered the values with deviation (in both directions but without outlier) from the average. The male/female ratio was also similar. The proposed first version of the algorithm was implemented for the classification of sleep stages of healthy adults. The tests also used adult subjects with no record of sleep disorder. In total, 27,662 time intervals were recorded, each with a length of thirty seconds. The classification matrices for the Charité subjects in sleep stages 4 and 3 are shown in the Tables 4 and 5. In the end, the results for the overall classification rate were: ● with 4 stages – accuracy: 58%, Cohen’s kappa: 0.50; ● with 3 stages – accuracy: 72%, Cohen’s kappa: 0.67. 2 Thomas Penzel, Dr. rer. medic. Martin Glos; initial study was carried out in Charité - Universitätsmedizin Berlin Center of Sleep Medicine Charitéplatz 1, D-10117 Berlin (Germany). Each row of the matrix contains the rates for both the classified and misclassified time intervals. For example, the first row in Table 4 indicates that 2649 intervals were correctly classified as the phase WAKE. The next entry displays the 1146 intervals classified as LS phase, but it was actually WAKE, and so on. Table 4: Classification results. Four stages. Stage (developed SW-System) WAKE LS DS REM Stage (expert) WAKE 2649 1146 480 219 LS 1686 10023 1700 942 DS 168 3041 2104 58 REM 628 1328 151 1339 Table 5: Classification results. Three stages. Stage (developed SW-System) WAKE NREM REM Stage (expert) WAKE 2472 1642 380 NREM 2060 16004 1658 REM 748 1101 1597 Tables 6 and 7 present the percentage of correct classification of sleep stages for each stage in particular. It can be seen, that in 4 stages, WAKE and Light Sleep is recognised with the highest accuracy, whereas the error rate at the Deep Sleep and REM stages is high, so an improvement in accuracy will be required. Table 6: Correctly classified stages. Four stages. Rights 58 % Mistakes 42 % WAKE 59 % LS 70 % DS 41 % REM 36 % Table 7: Correctly classified stages. Three stages Rights 72 % Mistakes 28 % WAKE 56 % NREM 81 % REM 42 % Discussion Supporting sleep professionals with the help of non-invasive system for long-term sleep study in the home environment is the goal of this research. In particular, this paper discusses the software part of the system. There are two main points, which represent the novelty of this work: the unique set of derived parameters which can be calculated from bio-vital signals and measured in a non-invasive way, and the use of multinomial logistical regression for the classification (which enables the possibility of easy algorithm adjustment – e.g. the consideration of transition patterns). The following paragraphs will elaborate on these points in more detail. To implement an automatic sleep stage analysis and evaluation in this project, the appropriate software was developed. This program had the task of automatically classifying a sleep phase in 30 second time intervals based on the vital parameters – respiration, heartbeat and body movement signals from raw PSG recordings. Moreover, for this technique to be transferable into later studies with the sensor set mentioned above, the signals from them would not be translated with the raw values of the recordings. The goal was to create an algorithm which would present the parameter signals through the relation to parameter’s average signal. The reason for the development of this method is to make evaluations which consider the differences in individuals. After the experiment was finished, the classification results were compared with the pre-estimated whole-night PSG recordings. The core of the software is the algorithm for sleep stage recognition using respiratory, heart rate and body movement signals. The results achieved so far are the following: 58% correct recognition rate for Wake, Light (N1 and N2), Deep (N3) and REM stages and up to a 72% hit rate for Wake, NREM and REM stages. Sleep recordings of 30 subjects were used for the evaluation. The outcomes obtained are promising, but improvements are still possible and they will be developed in order to obtain results which are closer to PSG application. There are several reasons for low accuracy of recognition in the REM stage. One of them is its similarity with WAKE sleep concerning breathing. Another more significant reason is that the REM stage is shorter than other sleep stages. Therefore, every incorrectly recognized epoch of REM sleep has percentagewise a high degree of influence on accuracy of recognition of this stage, whereas accuracy of the whole system will be less affected. Yet another reason for low accuracy of recognition is that the algorithm tends to recognise the next epoch as being the same as the previous one, if its features are similar. That being the case, since REM follows the LIGHT stage, the first “transitional” epochs are often misclassified due to the presence of similar characteristics (heart rate and breathing). The primary result of this project is that the selection of parameters (set of derived parameters from respiration, heart rate and body movement) has proved its validity. The theoretical foundation described in the chapter State of the Art has inspired this line of thinking and the results of this study have justified it. It is important to keep in mind, that there are still conceivable improvements to be made to the algorithm which could increase the accuracy of results, but would not change the main concept. Furthermore, these parameters can be obtained in a noninvasive way with a hardware system (e.g. with sensors placed under the bed’s mattress) that can be installed by non-experts, which would be important for a home sleep study system. Reducing the amount of derived parameters by using only 3 base features (heart rate, respiration rare and movement) will not lead to significant benefit for the proposed algorithm because the calculation of derived parameters does not need a high degree of computational power. To evaluate the introduction of respiration as a bio vital signal for sleep stage classification, the test was executed with the same dataset but without using the respiratory signal. The following results were achieved: general accuracy for WAKE/LS/DS/REM decreased to 53% and WAKE/NREM/REM to 64% (compared to 58% and 72% with respiratory signal). More importantly, the recognition rate of the REM stage diminished to about 15% for both sets of stages, which is not acceptable. The first attempt to use the balanced proportion of classes in the training dataset was made (WAKE/LIGHT/DEEP/REM = 25%/30%/25%/20%), but the total accuracy was decreased by 2% for WAKE/LIGHT/DEEP/REM and was not significantly changed for WAKE/NREM/REM stages. Nevertheless, the increase in the recognition rate of the REM stage has indicated the potential of this approach. The classification into three (Wake, NREM, REM) and four (Wake, Light, Deep, REM) sleep stages had to be performed according to the research goal. Multinomial logistical regression (MLR) is the classification method that was chosen here because it allows a categorical distribution of values. One dependent variable (in this case – sleep stages) has as input a set of different independent variables-parameters, and they may even overlap. More details are presented in the chapter Statistical Methods. Using MLR as a base for categorical distribution of sleep stage parameters will be used to enhance the process in the future, as it enables an easy and fast adjustment of the algorithm that depends on the requested input parameters. Even if just three bio vital signals are used in the method described, different derived parameters can be calculated when dependent variables are used in the regression model. Moreover, the adaptability of this algorithm allows for quick modifications if other bio vital signals are available for the study. Though the selection of suitable parameters for the model was one of the challenging tasks, it has led to satisfactory results. However, some improvements are still possible, for instance the identification of other derived parameters that can be obtained in a non-intrusive way. Selection of the most suitable parameters can be proposed on the basis of theoretical research, but proof is only possible after analysing the results of numerous studies. Even with a distinguished background in qualitative analysis, it would be necessary to test the proposed set of parameters with real data in order to confirm the theoretical assumptions. The reason for this is the complexity of biological processes in the human body which have a variety of effects on bio vital signals, which can be used as inputs for the algorithm. The consideration of transition patterns between different sleep stages was restricted to only one rule. Introducing other rules in respect to the transition matrix would increase the risk of reducing accuracy because if the previous sleep stage were incorrectly recognised, it would affect the classification of the next stages. The strategy of this approach has been confirmed as being a highly credible method for yielding successful sleep stage classification with the help of a greatly reduced set of parameters. The number of test subjects could be increased in order to collect and evaluate a higher amount of data. For this reason, further studies with different sets of input parameters and a higher number of subjects are in planning. During the study, several possible algorithm improvements were identified, and amongst others, the implementation of trust anchors will be investigated. The chapter Conclusions and Future Work focuses on this and other possibilities. Comparing the results of this work with other studies, it is important to keep in mind that input signals for the system presented here could be obtained in non-invasive way. Most of the scientific articles on sleep stage classification (e.g. [SUP17], [KAR18]) describe systems using Electroencephalography (EEG) as input. Using EEG can provide higher accuracy, but requires placing the sensors direct on the test subject’s head, which would be not acceptable for a non-invasive system. Therefore, the most similar approach to use for comparison would be ballistocardiography or other methods using respiration, heart rate and movement signals as input. In [MEN10] an accuracy rate of about 0,79 was achieved, but only the classification of NREM and REM stages was determined and WAKE stage was excluded. The research system proposed here also includes WAKE stage, which explains the slightly lower total accuracy, but nevertheless provides increased usefulness as a sleep stage identification system. Another paper [SAM14] describes an unobtrusive sleep stage classification system using a pressure-sensitive bed sheet which achieved 70% precision for WAKE, NREM and REM stages. With our system, we obtain a higher rate of accuracy and it was tested on a higher number of overnight PSG recordings (30 compared to 7). The number of features used in [LON14] is higher (14/13 compared to 10 in our method) and the results are equal without subject-specific normalization and just about 3% better with subject-specific normalization, but as mentioned before with a higher number of included features. In [AKT17] an accuracy rate of about 0.78 was achieved with a lower number of features. However, the proposed method can identify only SLEEP/WAKE stages, and in addition, an assumption was made to include NREM1 as a part of the WAKE stage. In the proposed in this paper method not only SLEEP/WAKE, but also WAKE/NREM/REM and WAKE/LIGHT/DEEP/REM classification is done, which is a different and much more complicated task. However, a comparison with the results from [AKT17] can be easily done using the already obtained results (having SLEEP=LIGHT+DEEP+REM). In this case even without adjusting the algorithm to recognise only SLEEP/WAKE stages, the approach presented in this paper got better results: accuracy: 0.84, sensitivity: 0.59, specificity: 0.89 and kappa: 0.46 compared to Acc=0.78, Se=0.85, Sp=0.48 and K=0.30 in [AKT17]. The fore-mentioned results and the abundant opportunities for advancement in sleep stage classification prove the potential of the concept presented here and the necessity for further research on this approach. The comparison with the state of the art articles has confirmed the quality of achieved results, having a better relation “amount of features” / “correctly classified sleep stages”. Conclusions and Further Work The results obtained in this study (72% accuracy, Cohen’s kappa – 0.67 for Wake, NREM and REM stages) prove that the algorithm developed for this project represents a truly promising method for sleep stage classification using just a few bio vital parameters: heart rate, respiration and body movement. It is also obvious that there are several ways to improve the classification algorithm. The first possibility is to define more derived parameters and examine them according to their significance in the MLR-model. For instance, the additional inclusion of sample entropy as an independent variable could potentially improve the model. Also, the classification data could be revised in a second step. If a particular category is predicted with a lower percentage, the time intervals directly preceding and following it could be re-analysed in order to improve the overall classification rate. Besides this, the intervals which were classified with a high-confidence level could act as trust anchors. They could adjust the percentage values for the category membership of their neighbours. Moreover, to further improve the software system, all the probabilities from the sleep stage transition matrix will be examined and two-step transitions [SCH14] will be introduced. Furthermore, an investigation is planned to study the effect of a currently unbalanced number of samples of different sleep stages on the accuracy of the method [TIM09]. One of the possible solutions could be to consider the balanced proportion of classes (WAKE, NREM, REM) for training dataset with executing of following test classification, but this point should be deeply investigated, as the first results have indicated. The reduction of used derived features is also planned for the future when executing studies with higher numbers of test subjects would be possible. Another approach to the classification of sleep phases is the implementation of a multi-layer neural network algorithm with the MLR model in the first layer, though it is still questionable whether the neural network analysis would improve the classification rates. Thus, it should be considered, but thoroughly examined and scrutinized. A perfect match of the evaluated results with the findings of a sleep medicine professional is not possible due to the R-K method’s judgment standards, which include ambiguities. However, all in all it is still desirable to develop a more accurate and, more importantly, non-invasive solution for the problem of recording and estimating the sleep stages. It would help to find alternative methods which could, in time, substitute for the more complex PSG procedure. This research is part of an intended far larger study aiming to develop a monitoring system for home-based sleep analysis. It was particularly focused on designing a software solution that can classify the sleep stages. As mentioned earlier, the ultimate goal is to obtain the signals of particular bio vital parameters via a set of pressure sensitive sensors placed under the mattresses of study subjects. This approach should presumably ensure that the sleep recording process does not disturb the sleep of the study participants. Furthermore, to get proper recordings of a natural sleep state, the subjects must feel as though they are sleeping in a normal bed in familiar conditions. The planned system will include both software and hardware parts. The hardware portion will provide data collection and pre-processing. The data recording will be done through a set of sensors and necessary filters will be applied directly to the hardware to record movement, breathing and heart rate signals. After that the information will be automatically transferred to a server, where the software will take care of further operations. It will be processed there and analysed by the developed sleep stage algorithm. The interpretation of the acquired results has to be made by medical professionals. After completing the development and integration of all the system parts, the resulting monitoring system could provide useful diagnostic support to sleep professionals and enhance the field of sleep medicine by offering a convenient way to perform preliminary sleep studies in the home environment. Acknowledgements We thank the Interdisciplinary Center for Sleep Medicine of Charité Clinic in Berlin and in particular Dr. rer. medic. Martin Glos, Beate Diecker and Claudia Biro for supporting the study. We would also like to thank Agnes Klein for her help, which was very important for the development of the system. This research was partially funded by the EU Interreg V-Program "Alpenrhein-Bodensee-Hochrhein": Project "IBH Living Lab Active and Assisted Living", grants ABH40 and ABH66. References [AKT17] Aktaruzzaman Md. et al: “Performance comparison between wrist and chest actigraphy in combination with heart rate variability for sleep classification”, Computers in Biology and Medicine, 2017. [AVE98] Alexander Von Eye, Christof S.: “Regression Analysis for Social Sciences”, Academic Press, 1998. [BEL91] Belsley D. A.: “Conditioning Diagnostics: Collinearity and Weak Data in Regression”, Wiley- Interscience, 1991 (Wiley Series in Probability and Statistics (Book 262)). [BER15] Berry R.B., Wagner M.H.: “Indications for Polysomnography, Portable Monitoring, and Actigraphy”, Sleep Medicine Pearls (Third Edition), 2015. [BOC18] Bochkanov S.: “ALGLIB Project”. http://www.alglib.net/dataanalysis/logit.php. Version: 2018, accessed: September 28, 2018. [BR18] Brain Ripley, William V.: “Package ‘nnet’”, https://cran.rproject.org/web/packages/nnet/nnet.pdf. Version: 2017, accessed: June 14, 2018. [CHE03] Chen H.C., Chen S.W.: “A moving average based filtering system with its application to real-time QRS detection”, Computers in Cardiology, 2003, pp. 585–588. [CHO07] Choi B. H., Choi-H. J. M., Shin B., Lee J. Y., Jeong K. D. U., Seo S. J. W., Seo J. W.: “Nonconstraining sleep/wake monitoring system using bed actigraphy”, Medical & biological engineering & computing 45, 2007, pp. 107–114. [CHO13] Chokroverty S., Thomas R.: “Atlas of Sleep Medicine” 2nd edition, Elsevier Ltd. Oxford, 2013. [CHU07] Chung G.S., Kim K.K., Lim Y.G., Choi J.W., Jeong D-U., Park K.S., Choi B.H.: “REM Sleep Classification with Respiration Rates”, Tokyo: IEEE, November 2007, pp. 194–197. [DCM12] Douglas C. M., Geoffrey V.G., Peck E.A.: “Introduction to Linear Regression Analysis” 5th edition, Wiley, 2012. [ETT10] Ettinger S.J., Feldman E.C.: “Textbook of Veterinary Internal Medicine Expert Consult”, Expert Consult. 2nd edition, 2010. [GAI18] Gaiduk M., Vunderl B., Seepold R., Ortega J.A., Penzel Th.: “Sensor-Mesh-Based System with Application on Sleep Study”, Bioinformatics and Biomedical Engineering IWBBIO 2018, Lecture Notes in Computer Science, vol 10814. Springer, Cham, 2018. [GAR16] Garson G. D.: “Logistic Regression: Binary & Multinomial: 2016 Edition”, Statistical Associates Publishers, 2016. [GBH06] Guang-Bin H., Chee-Kheong S., Qin-Yu Z. Z., Qin-Yu Z.: “Extreme learning machine: Theory and applications”, Neurocomputing 70, Singapore: ScienceDirect, 2006, pp. 489–501. [HAY12] Hayet W., Slim Y.: “Sleep-Wake Stages Classification Based on Heart Rate Variability”, International Conference on BioMedical Engineering and Informatics (BMEI 2012), IEEE, 2012, pp. 996–999. [KAM12] Kambayashi Y., Hagiwara H.: “Estimating Sleep Cycle Using Body Movement Density”, Biomedical Engineering and Informatics (BMEI), IEEE, 2012, pp. 1081–1085. [KAR18] Karimzadeh F., Boostani R., Seraj E. and Sameni R.: "A Distributed Classification Procedure for Automatic Sleep Stage Scoring Based on Instantaneous Electroencephalogram Phase and Envelope Features,", IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 26, no. 2, pp. 362-370, Feb. 2018. [KAT11] Katz M. H.: “Multivariable Analysis: A Practical Guide for Clinicians and Public Health Researchers”, 3rd edition, Cambridge University Press, 2011. [KUR10] Kurihara Y., Hiroshi T., Watanabe K.: “Sleep-States-Transition Model by Body Movement and Estimation of Sleep-Stage-Appearance Probabilities by Kalman Filter”, IEEE Transactions on Information Technology in Biomedicine 14, 2010, Nr. 6, pp. 1428–1435. [KUR12] Kurihara Y., Watanabe K.: “Sleep-Stage Decision Algorithm by Using Heartbeat and Body- Movement Signals”, IEEE Transactions on Systems, Man, and Cybernetics - Part A: Systems and Humans, IEEE, 2012, pp. 1450 – 1459. [LEB01] Le Bon O., Hoffmann G., Dramaix M., San Sebastian I., Murphy J. R., Kentos M., Pelc I., Staner P. L. L., Staner S. L.: “The first-night effect may last more than one night”, Journal of Psychiatric Research 35, 2001, Nr. 3, pp. 165–172. [LEE12] Lee-Chiong T.L., Brooks R., Mattice C.: “Fundamentals of Sleep Technology”, Lippincott Williams & Wilkins, 2012. [LON14] Long X., Fonseca P., Haakma R., Aarts R.M., Foussier J.: “Analyzing respiratory effort amplitude for automated sleep stage classification”, Biomedical Signal Processing and Control 14, 2014, pp. 197–205. [MEN10] Mendez M.O., Matteucci M., Castronovo V., Ferini-Strambi L., Cerutti S. and Bianchi A.M.: “Sleep staging from heart rate variability: time-varying spectral features and hidden Markov models”, Int. J. Biomed. Eng. Technol. 3. 2010. [NAT11] National Heart Lung and Blood Institute (NHLBI): “Your Guide to Healthy Sleep”, NIH Publication No. 11-5271, August 2011. [PEN16] Penzel Th., Kantelhardt J.W., Bartsch R.P., Riedl M., Kraemer J.F., Wessel N., Garcia C., Glos M., Fietze I., Schöbel C.: “Modulations of Heart Rate, ECG, and Cardio-Respiratory Coupling Observed in Polysomnography”, Front Physiol, 2016, doi:10.3389/fphys.2016.00460. [RF17] R Foundation: “The R Project for Statistical Computing”, https://www.r-project.org/. Version: 2017, accessed: March 19, 2018. [SAM14] Samy L., Huang M.C., Liu J.J., Xu W. and Sarrafzadeh M.: "Unobtrusive Sleep Stage Identification Using a Pressure-Sensitive Bed Sheet," in IEEE Sensors Journal, vol. 14, no. 7, pp. 2092-2101, July 2014. [SCH14] Schlemmer A., Parlitz U., Luther S., Wessel N., Penzel Th.: “Changes of sleep-stage transitions due to ageing and sleep disorder”, Phil. Trans. R. Soc., doi: 10.1098/rsta.2014.0093, published 29.12.2014. [SIN15] Singh J., Badr M.S., Diebert W. et al: “American Academy of Sleep Medicine (AASM) Position Paper for the Use of Telemedicine for the Diagnosis and Treatment of Sleep Disorders: An American Academy of Sleep Medicine Position Paper.”, Journal of Clinical Sleep Medicine : JCSM : Official Publication of the American Academy of Sleep Medicine 11.10 (2015): 1187–1198, doi:10.5664/jcsm.5098. [SPR09] Spriggs W. H.: “Essentials of Polysomnography”, Jones & Bartlett Learning, 2009. [SUP17] Supratak A., Dong H., Wu C. and Guo Y.: "DeepSleepNet: A Model for Automatic Sleep Stage Scoring Based on Raw Single-Channel EEG,", IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 25, no. 11, pp. 1998-2008, 2017. [TAT15] Tataraidze A., Anishchenko L., Korostovtseva L., Kooij B.J., Bochkarev M. and Sviryaev Y.: "Sleep stage classification based on respiratory signal," 2015 37th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), Milan, 2015, pp. 358-361. doi: 10.1109/EMBC.2015.7318373. [TIM09] Tilmanne J., Urbain J., Kothare M.V., Wouwer A.V., Kothare S.V.: “Algorithms for sleep-wake identification using actigraphy: a comparative study and new results”, Journal of Sleep Research, Volume 18, Issue 1, pp. 85-98, 2009. [UCL18] UCLA Institute for Digital Research and Education: “R Data Analysis Examples: Multinomial Logistic Regression”, http://www.ats.ucla.edu/stat/r/dae/mlogit.htm, accessed: March 19, 2018. [YOO17] Yoon H., Hwang S.H., Choi J-W., Lee Y.J., Jeong D-U. and Park K.S.: “REM sleep estimation based on autonomic dynamics using R–R intervals”, Physiological Measurement, Volume 38, Number 4, 2017. 45 Digital Health and Care Study on Elderly Monitoring Overview The paper addresses the problem of designing sustainable home healthcare systems for the elderly. Particular focus is given to sleep analysis as one of the important parts of such systems. Context This publication was prefaced by multiple scientific publications in the area of “Healthcare Technologies” and active participation in several international projects on the given topic. Practical experiences from the projects and knowledge obtained from in-depth research have resulted in this publication. The findings of a study carried out within the framework of the international project "IBH AAL: Home Health Living Lab" were evaluated and scientifically supported. The research stays in Ancona (Italy) and Seville (Spain), among others, have provided essential knowledge that has been crucial in the implementation of this research. Journal information A Special Issue “Sustainable Technology and Elderly Life” of the Journal “Sustainability” was selected for the submission of this paper due to the relevance of the topic “Elderly Healthcare”. This journal is indexed in JCR with the current Impact Factor of 3.251, and its ranking for the last five years is presented in the following Table. Category Environmental Sciences Environmental Studies Green & Sustainable Science & Technology JCR Year JIF Rank JIF Quartile JIF Rank JIF Quartile JIF Rank JIF Quartile 2020 124/274 Q2 59/125 Q2 6/9 Q3 2019 120/265 Q2 53/123 Q2 6/8 Q3 2018 105/251 Q2 44/116 Q2 3/6 Q2 2017 121/242 Q2 51/109 Q2 3/6 Q2 2016 119/229 Q2 47/105 Q2 4/6 Q3 Sustainability 2021,13, 13376 6 of 18 circles and has undergone multiple evaluations [ 43 , 44 ]. We used a German version of the questionnaire proposed by reference [45]. Standardization in the true sense does not exist for the PSQI. PSQI classification results from the cut-off value of 5; it was calculated in the original work [ 42 ] based on the classification of people with sleep disorders and healthy individuals. A total of 18 items are used for quantitative evaluation. They were assigned to seven components, each assuming a value range from 0 to 3. The total score results from the summation of the component scores and can range from 0 to 21. A higher score corresponds to a lower quality of sleep. There is an empirically determined cut-off value (of 5) that allows a division into “good” and “bad” sleepers. A representative study for the German-speaking area [ 46 ] surveyed 1049 participants. Here, a proportion of 32.1% of the participants had a PSQI total score > 5. This survey might be used to compare the findings within our study. 2.3. Acceptance of Technologies by the Elderly Since the acceptance of technologies plays an essential role in their widespread use, this aspect was addressed in the study. Based on the literature review, several points were selected to be considered during the implementation period in order to enable an evaluation afterward: • The technologies should be self-explanatory and should not require extensive training. •The devices should function automatically as much as possible; •The devices should be comfortable and safe to use. To allow a comprehensive analysis, the following methods were selected: •Surveying with a questionnaire. •Free conversations with test participants to receive unstructured feedback. •Systematic analysis of occurrences and irregularities during the study. In preparing the questionnaire, an emphasis was placed on ease of use, safety during use, and readiness and suitability of the technologies for regular use. It was also planned to analyse whether there were any difficulties with the use of the devices and the reasons for those difficulties. For this purpose, in addition to answering the standardised questions, the test subjects were asked to provide a free comment (related to the respective device) in case of irregularities/problems with one of the devices being used. Questions were asked during the visits that took place over the 14 days and on the last day, together with the final questionnaire. To evaluate the acceptance of technologies for different health-related concerns, the system was used that not only included the EmFit QS+ sleep analysis device described above but also included the Polar OH1 heart rate monitor. This kind of device (photoplethysmography based on PPG) can be applied to measure stress levels, which was not performed as part of the presented study. Currently, heart rate variability is often used to detect stress [ 47 , 48 ]. However, existing studies have shown that heart rate variability may be often substituted by pulse rate variability (which can be obtained by PPG measurement), especially at rest [ 49 ]. Moreover, some studies have presented the possibility of stress detection by analysing pulse rate variability [ 50 ]. However, it is important to note that the Polar OH1 device in this study was used exclusively for qualitative and not quantitative measurement and should only represent a wearable that should be placed on the upper or lower arm and that could be used for stress measurement after appropriate analysis. During the study, the OH1 was to be placed on the right upper or lower arm daily for a fortnight. When the sensor was close to the provided smartphone with the app installed and when new data were available, it paired with the application, and the data were collected for future analysis. The test subjects only had to place the sensor on the upper arm and turn it on by pressing a button. It was also necessary to charge the sensors at night. No other actions were required from the test subjects. Sustainability 2021,13, 13376 7 of 18 3. Results After the quantitative analysis of the sleep measurement as well as the qualitative analysis of the devices for both the sleep and heart rate measurements were conducted, a set of results was obtained, which are presented below. The difference between subjective and objective sleep quality measurement was analysed. Figure 2shows a corresponding box plot diagram for the ten participants. The vertical axis represents the average differences between subjective and objective measurement for all of the subjects who participated in the study. We can see a clear tendency to underestimate subjective sleep quality compared to the value measured with the electronic device. This underestimation remains relatively stable, with a median value of approximately 13% and a mean value of approximately 10%. However, there are also individuals for whom the difference between objectively and subjectively measured sleep quality is notably more significant. This fact should be taken into account when planning the use of technologies to measure sleep quality. For two subjects, there are significant differences in the number of objective and subjective measurements available. For these subjects in particular, there can be discrepancies in the evaluation of the differences between the subjective and objective measurement, which can also lead to some variation in the results. With the total number of available recordings (115–objective and 90–subjective), it is nevertheless to be assumed that the results are also transferable to the overall population, which is also confirmed below with the calculation of margin of error. Figure 2. Analysis of the differences between objective and subjective measures of sleep quality for ten subjects. Negative values—subjective sleep quality is underestimated. To give an idea of the possible differences between objective and subjective measures of sleep quality, the comparison of the measured values (possible values are in the interval from 1 to 10) for a subject with the most significant differences participating in the study is shown in Figure 3. It can be seen that the median value of sleep quality for the total two-week study period is more than twice as high for the objective measurements taken with the help of an electronic device (approximately 9.2 points of sleep quality) than it is for the subjective Sustainability 2021,13, 13376 8 of 18 measurements taken with the questionnaire (approximately four points of sleep quality). When analysing these results, one can identify two main reasons for this difference: firstly, if there is a large difference between the number of recordings available for both (objective and subjective) measurement methods, then there is more a likelihood that the difference that is measured is greater; secondly, since different parameters are measured in the objective and subjective measurements, the perceived sleep quality may differ significantly from the measured one because measurement with a device cannot take into account all aspects that have an impact on human perception. People typically do not divide sleep quality into the sum of parameters but perceive an overall impression. Figure 3. Comparison of subjective and objective measurements for a subject with the most significant differences between two measurement methods. The results of the PSQI survey are shown in Table 1. From this, it can be seen that there were significantly more people identified themselves as “bad” sleepers than in the previous study reported in reference [46]. There may be several reasons for this, such as: • The study by reference [ 46 ] analysed a broad demographic. In our study, only subjects over 65 years of age were included. Therefore, the results may differ significantly due to the different age groups. • About 20 years has passed between the study conducted by reference [ 46 ] and our study. During this period, the prevalence of sleep-related disorders may have been changed. Table 1. PSQI screening results. PSQI Value Percentage ≤5 40% >5 60% The results of the interviews that were conducted at the end of the study are presented in box plots in Figures 4and 5. Furthermore, the results are analysed and discussed in the following text. As a result of the analysis of the box plots in Figure 4, the following conclusions can be drawn: • There was no agreement among test participants on whether they would use the devices regularly. It should be noted that the devices were used over a two-week period, which, among other things, means that the subjects did not have long-term experience with the devices and therefore could not assess whether they would use the devices regularly in the long term well. Sustainability 2021,13, 13376 9 of 18 • The questions regarding the complexity of the devices, ease of use, and the need for support to use the devices, there is a clear trend in the answers, which is more in favour of the devices being relatively easy to use independently without external support. Moreover, despite the novelty of these types of devices for the subjects, no unnecessary complexity was perceived. According to the technology acceptance model, these points indicate that the proposed concept can increase acceptance [51]. Figure 4. Box plot of the interview results. 1—totally disagree, 5—strongly agree: ( a ) I can imagine using the devices regularly. ( b ) I find the devices to be unnecessarily complicated. ( c ) I find the devices easy to use. (d) I think I would need technical support to use the devices. Figure 5. Box plot of the interview results. 1—totally disagree, 5—strongly agree: ( e ) I find that there are too many inconsistencies in the devices. ( f ) I can imagine that most people learn to handle the devices quickly. ( g ) I find the operating of the devices very complicated. ( h ) I felt very safe using the devices. (i) I had to learn many things before I could handle the devices. Sustainability 2021,13, 13376 10 of 18 Analysing Figure 5, one can see that at several points, an obvious conclusion can be drawn that: • The subjects do not believe that there are too many inconsistencies (confusing or unclear functions or components) with the devices utilized. • According to the subjects’ opinions, the majority of people can quickly learn how to handle the devices. • The operation of the devices is not very complicated according to the subjective perception of the test participants. However, it should be noted that several people noticed some irregularities in the functioning of the proposed devices, which are described in detail below. • There was no concern about safety while using the equipment. This means that the perceived risk of using the technologies was relatively low, which according to references [52,53], is essential for the acceptance of technologies. • It was possible to use the devices without having to learn much new information beforehand. In summary, from the answers to question (e), question (b), and question (c) , it can be concluded that the presented concept, where the subjects had to interact with the devices as little as possible, gives a sense of simplicity to users, which can also lead to an increase in the acceptance of such systems [54–56]. When interpreting the results obtained from the surveys, it can be said that the twoweek study periods provided the test subjects with sufficient information to be able to assess the ease and safety of using the devices, which can be seen from the fairly clear answers. On the other hand, the strong variance and tendency towards “not sure” in the answers to the question of whether the test participants could imagine using the devices regularly shows us that a significantly longer period of use or familiarization is necessary in order to obtain a clear answer to this question. Another point that can be observed is the fact that there is at least one outlier in the majority of answers to the questions. There may be different reasons for this. One of them is that it is possible that some questions were misunderstood, resulting in a reverse answer being given. For example, instead of “strongly agree”, the answer becomes “strong disagree”. This could be caused by the fact that some questions deliberately include two variants in questionnaires to make sure that people answer the questions consciously, the so-called control mechanism. In individual cases, it can lead to misunderstandings or show us that people answer some questions without understanding them precisely. In general, however, it helps to exclude the questions that are not carefully considered by the participants before answering to be excluded from the analysis. Another explanation for the presence of the outliers could be the fact that the individual rare occurrences with the devices, which only happened to a single person, led to this being very distinct from the majority opinion. This could be, for example, a technical malfunction that only occurs once to one person. A total of 10 participants participated in the study due to the targeting of a specific group of users and the use of not only questionnaires but also the use and installation of the hardware for several devices to be used over a fortnight. Considering this fact, the question arises as to whether the difference to the expected results by the entire population would differ greatly. This point needs to be subject to deep scientific discussion in order to make a well-founded statement. In the following, this question is analysed and supported by statistical evaluation. To forecast a maximum possible deviation between the results of a study carried out with a sample (ten people in our case) and the entire population, the parameter called “margin of error” is typically used. We have also followed this scientifically recognized approach. According to references [ 57 , 58 ], we calculated the value of the margin of error for each question. For that, (2) was used: d=trN−n N−1rS2 n(2) Sustainability 2021,13, 13376 11 of 18 where tis the Student’s t-critical value for normal distribution, N is the size of the population, n is the sample size, and S is the population standard deviation. Considering the fact that the population size is substantially larger than the used sample size, qN−n N−1 tends towards «1». Due to the impossibility of calculating the standard deviation of the entire population, Scan be replaced by s(standard deviation of the sample). Taking into account the mentioned factors, Equation (1) can be transformed as follows: d=ts √n(3) The tvalue for the sample size of 10 with 9 degrees of freedom and for the significance level α = 0.05 (for the confidence level of 95%) can be found in the corresponding table and is equal to ± 2.262156 (two-tailed). Knowing this, the margin of error for every question was calculated with Equation (2). The results with a confidence level of 95% are presented in Table 2. The possible range of a mean value for the entire population according to the calculation of the margin of error is represented in Figures 4and 5, with a coloured rectangle representing each question. Analysing that, even in the case of the maximum possible deviation of the mean value for the entire population from the sample mean according to margin of error calculation, for all of the interview questions except for “I can imagine using the devices regularly”, the values still indicate clear agreement or disagreement with the corresponding question, which was also the case for the sample from the current study. This is due to the fact that the responses are very homogeneous and provide us with a clear picture, even with the available number of test persons. Out of that, taking into account the performed margin of error calculation, the results of the study conducted with 10 subjects may be extrapolated to an entire elderly population and can provide a significant scientific added value due to the clear and explicit distribution of the responses. Therefore, it can be stated that no meaningful difference in the results is expected in cases where the study is performed with a large sample size, even for the entire population. Table 2. Margin of error calculation. Measure Question (a) (b) (c) (d) (e) (f) (g) (h) (i) SD 1.6865 1.3375 0.8498 1.4944 0.0000 0.8433 1.2649 0.6325 0.3162 Mean 3.2000 1.7000 4.5000 1.7000 1.0000 4.6000 1.6000 4.8000 1.1000 Sample standard error 0.5333 0.4230 0.2687 0.4726 0.0000 0.2667 0.4000 0.2000 0.1000 Margin of error 1.2065 0.9568 0.6079 1.0691 0.0000 0.6032 0.9049 0.4524 0.2262 Similarly, the margin of error value can be calculated for the obtained difference between objective and subjective measurement. Performing the calculations according to (3), the margin of error for this measurement with 10 subjects is equal to 1.35, which means that for a 95% confidence level, the mean value of difference for the entire elderly population would be within the interval − 2.37–0.33 ( − 1.02 for the sample of 10 subjects). This means that even in the case of the maximal possible deviation of the mean value of the difference between the objective and subjective measurement, there will still be a clear correlation between these two methods of measurement for the entire population, and the difference will be between 3.3% and 23.7% (0.7–21.1% for a 90% confidence level). This allows one to extrapolate the results of the study conducted with 10 subjects to the entire elderly population with the confidence level mentioned and by considering possible deviations. Based on a statistical evaluation performed using the margin of error calculation, the study results with 10 subjects provide high qualitative and scientifically significant results. It is also important to mention that although only 10 subjects participated in the study, the total number of test nights for the calculation of the difference between subjective and objective measurements was equal to 140, which is a significant number and also increases the reliability of the results that were obtained, as the exact number of evaluation nights Sustainability 2021,13, 13376 12 of 18 (and not only the number of subjects) is relevant for the calculation of the differences between the two types of measurement. Although the survey results have shown that test persons see the proposed technologies as easy to use, some irregularities can be observed when analysing the available recordings. As shown in Table 3, neither for the objective measurement nor for the subjective measurement of sleep quality are all recordings for all 14 days of the study available. Table 3. Availability of measurements per person (study duration—14 days). Measurement Subject 12345678910 Subjective 8 10 14 9 11 13 7 7 5 6 Objective 7 6 14 14 14 14 9 10 13 14 If we visualize the number of available recordings as in Figure 6and analyse them, we can see significantly more recordings with the objective method. This is because the test participants did not have to make any extra effort. After all, the device automatically took recordings. In the case of subjective measurement, the subjects had to tag a perceived sleep quality on their own. However, even in terms of the objective measurements, there were some days where there were no recordings. According to the interviews that were conducted, there were two main reasons for this: the subject did not sleep in the bed that night, or the device was turned off from the power socket, which was not necessary but allowed. Sometimes, as the subjects self-reported, they forgot to plug the device back into the socket before going to bed. Figure 6. Box plot representing the number of days with available measurements for all subjects for both approaches. There was only one person who filled out the questionnaire on all 14 days. All of the device recordings were also available for this person. Furthermore, although the questionnaire was created, there was an attempt to make it as simple and as motivating as possible. That is why the graphic form was chosen. It follows that the more effort a method requires from the user (even if it is minimal), the more incompleteness there will be in the data collected. Sustainability 2021,13, 13376 13 of 18 It is important to address the fact that there is a certain imbalance between the number of objective and subjective measurements in order to assess whether this led to a distortion of the results when evaluating the differences between the objective and subjective measurements presented above. We have a total of 115 nights for the objective measurement and 90 nights for the subjective measurement, which corresponds to a difference of only about 20%. For the majority of people, there are more objective measurements than subjective ones. However, it is important to note that there are also people who have more subjective measurements than objective ones, which means that the differences in evaluation can exist in both directions, but this affects the overall evaluation less than a one-sided shift in the number of measurements. Since the total number of nights is significant (as there is an imbalance exactly in the number of night measurements and not in the number of subjects) in our study and since the evaluation is already statistically significant both at 90 and 115 nights independently, it can be assumed that this imbalance only plays a minor role in the overall evaluation. Another important point is that the difference in the number of subjective/objective measurements is relatively small for most subjects. Only in two subjects (9 and 10) are significant differences present in the number of objective and subjective measurements. These two subjects are mainly responsible for the imbalance (18 out of 25 nights of total difference). With these subjects in particular, there are indeed discrepancies in the evaluation. However, the overall evaluation should remain statistically significant. Another important finding is that 50% of the test subjects noticed anomalies in the functioning of the home health devices. After a detailed analysis of these reports, it turned out that the main reason for this was the Polar OH1 heart rate measurement device. There were two main problems with it: • Firstly, the switching button was tiny and partially placed in the device’s body. This made it difficult for the subjects to press the button, especially when their fine motor skills were not perfect. • The second problem was the lack of direct feedback from the device. Therefore, the test subjects did not immediately recognize whether the device was already switched on or not. After evaluating the test results, we found that several recordings were only a few seconds long. This means that the subjects switched on the device, but because they were unsure whether it worked, they tried to switch on the device again, which eventually led to the device being turned off. It is important to note that the evaluation mainly tried to cover the general points that would also be transferable to other technologies, e.g., the automatic functioning of the system without the need of user actions or the need for feedback from the device will be equally relevant for other technologies. Nevertheless, it is not excluded that, in case of using another system, there may be some differences, which are also listed further in limitations. 4. Conclusions The study allowed us to collect and analyse relevant data on the use of technologies for older adults. Several conclusions that provide new information to the scientific community can be drawn from the evaluation and are presented below. One of the questions to be addressed in the study framework was whether an objective measurement with an electronic device could replace a subjective measurement of sleep quality, which is typically conducted through the use of a questionnaire. The analysis that was carried out confirms the possibility of this substitution when it comes to evaluating the sleep quality of a group of people and when the overall average result is of interest. For example, it could be used to conduct a comprehensive study of sleep quality in a region or among a specific subset of people. Considering the accuracy obtained for each person, a substitution is not recommended because it has been shown that the differences between the two measurement methods may be extreme in some infrequent exceptional cases. It is important to note that the results of the study do not mean that a subjective measurement Sustainability 2021,13, 13376 14 of 18 will produce more accurate results. It only means that in some people, the results will differ significantly, and in the practice of sleep medicine, a subjective measurement of several sleep parameters has been the most commonly used method (e.g., for the detection of insomnia [ 59 ]). Therefore, this method is a standard procedure, and because it has been used for such a long time, health care professionals have a great deal of experience in evaluating and interpreting these kinds of measurements. For some other measurements, objective measurement methods are used in sleep medicine, for example, for the detection of sleep phases [ 60 ]. Therefore, a combination of subjective and objective measurements can currently be recommended to obtain a comprehensive analysis of sleep that includes its different characteristics. Another critical point examined in the study is whether the elderly are comfortable with the existing technologies for measuring health-related parameters and which aspects should be considered when implementing home health systems. The concept of implementing the technologies created for the study had a few key points: • The use of devices that required minimal action on the part of the users. For example, no actions are required except (voluntary) for the unplugging and subsequent replugging of the sleep analysis device into the power socket. In the case of the device measuring heart rate during the day, only pressing a button was necessary, and the evaluation of the study showed that even this minimal necessary action led to problems. A possible solution would be to use a device with a more prominent and easier-to-use button to turn it on. • The technologies should be self-explanatory, and no complex training or support from a third person should be necessary. This goal was achieved according to the results of the interview, as explained in detail below. • When using electronic devices, there should be a sense of safety. This means that the devices should not contain any parts that could be considered dangerous. Additionally, users should be assured that when using the technologies, they cannot be easily broken. For this, the devices should be robust enough and have as few as possible easily breakable parts. Another point that was not considered when planning the concept, but which became apparent during the evaluation, is that direct feedback from the devices to the user would increase the feeling that the device was working correctly. This would also help to avoid certain operating errors. The experiences during the study are in line with the usability heuristics known from reference [61]. A significant result was obtained when subjectively measuring sleep quality with PSQI—60 % of the test subjects were identified as “bad” sleepers. This clearly shows that the broad use of sleep monitoring technologies would enhance early diagnosis because many test subjects were not even aware that they might have sleep problems that could be treated. Additionally, currently, when the constant development of technologies and algorithms allows new and less invasive methods to detect sleep stages [ 62 ], it is possible to make this early detection easier and more convenient for users. By directly comparing objective and subjective measurement methods, it can be seen that in the case of automatic measurement using technologies, it is possible to expect better data completeness and consistency than the manual filling out of questionnaires. This may ultimately lead to more relevant data being collected and analysed early, improving the quality of life and possibly the users’ health. The test subjects used the provided devices for a fortnight and finally gave their assessment during an interview. The results allowed us to conclude that the proposed concept is a successful model in terms of the simplicity of independent device operation o and a sense of security when using them. However, the test subjects were not sure as to whether they would use the devices regularly. One explanation for this is that the benefits of using these devices are not always clear to the subjects—there was feedback that they did not think that using these devices would improve their quality of life. To Sustainability 2021,13, 13376 15 of 18 overcome this barrier, a sustained educational effort is needed to communicate clearly and understandably the potential benefits the elderly population using such types of technology. Without this development of motivation to use the technologies, widespread use is complicated, even though handling is not necessarily a big problem, as the conducted study has shown. There are some limitations in the work presented: • Only a few technologies could be used and evaluated within the study framework. Therefore, it cannot be excluded that the results could deviate with a different selection of devices. However, it is always necessary to select a specific subset of technologies because it is impossible to test all available devices at once. • The number of test subjects was limited to 10. In order to obtain statistically relevant data, the period of 14 days was chosen for the study, resulting in a total of 140 persondays of study. In addition, the same proportion of male and female test subjects was ensured. Furthermore, the proposed study contained a certain type of usability testing, which included questionnaires and a free-form interview to understand if there were any problems with the usability of devices. According to reference [ 63 ], only 10 ±2 subjects are necessary to discover 80% of usability problems. Moreover, the analysis of the transferability of the results to the entire population of persons aged 65 and older with the approach of the statistical parameter “margin of error” was carried out and presented in the section “Results”. The margin of error scientifically confirmed the significance of the results that were obtained. • Since the study target group was people aged 65 years of age and older, the results cannot be directly transferred to other age groups. • The technologies were not connected to a common platform that could be accessed by the participants directly. Therefore, it was impossible to assess whether the possibility of directly viewing the results of the recordings by the test subjects could have a positive impact on acceptance. To overcome the limitations mentioned above and to gather additional valuable information, the next step could be planning and organizing a new extended study. This could involve more subjects and could allow a wider range of technologies connected to a common platform to be tested. Furthermore, further work could be conducted on the selection and design of questionnaires for the collection of relevant data for analysis. Author Contributions: Conceptualization, M.G., R.S. and N.M.M.; methodology, M.G.; validation, M.G; formal analysis, M.G.; investigation, M.G.; resources, M.G. and R.S.; data curation, M.G.; writing—original draft preparation, M.G.; writing—review and editing, R.S., N.M.M. and J.A.O.; visualization, M.G.; supervision, R.S. and J.A.O.; project administration, R.S., M.G.; funding acquisition, R.S. All authors have read and agreed to the published version of the manuscript. Funding: This research was funded by the EU Interreg V-Program “Alpenrhein-Bodensee-Hochrhein”: Project “IBH Living Lab Active and Assisted Living”, grants ABH40, ABH41, and ABH66. This research was also partially funded by the Carl Zeiss Foundation (project number: P2019-03-003), by the German Federal Ministry for Economic Affairs and Energy, ZiM project “Sleep Lab at Home” (SLaH) grant: ZF4825301AW9 and grant PGC2018-102145-B-C21, funded by MCIN/AEI/ 10.13039/501100011033 and by “ERDF A way of making Europe”, funded by the Ministry of Science, Research and the Arts Baden-Württemberg and supported by the Open Access Publication Fund of the HTWG Konstanz University of Applied Sciences. Institutional Review Board Statement: Ethical review and approval were waived for this study, based on the review by ethics officers from the HTWG Konstanz and the University of Applied Sciences Kempten, Germany. Informed Consent Statement: Informed consent was obtained from all subjects involved in the study. Acknowledgments: We thank the caregivers of the Workers’ Welfare Association (AWO) Schwarzwald- Baar e.V. for their contribution in conducting the study. Conflicts of Interest: The authors declare no conflict of interest. 68 M. Gaiduk et al.: Estimation of Sleep Stages Analyzing Respiratory and Movement Signals 1 Abstract— The scoring of sleep stages is an essential part of sleep studies. The main objective of this research is to provide an algorithm for the automatic classification of sleep stages using signals that may be obtained in a nonobtrusive way. After reviewing the relevant research, the authors selected a multinomial logistic regression as the basis for their approach. Several parameters were derived from movement and breathing signals, and their combinations were investigated to develop an accurate and stable algorithm. The algorithm was implemented to produce successful results: the accuracy of the recognition of Wake/NREM/REM stages is equal to 73%, with Cohen's kappa of 0.44 for the analyzed 19324 sleep epochs of 30 seconds each. This approach has the advantage of using the only movement and breathing signals, which can be recorded with less effort than heart or brainwave signals, and requiring only four derived parameters for the calculations. Therefore, the new system is a significant improvement for non-obtrusive sleep stage identification compared to existing approaches. Index Terms—biomedical signal processing, regression analysis, sleep stages, sleep study. I. INTRODUCTION EVERAL studies show the importance of sleep for maintaining good health [1, 2, 3]. They emphasize its duration as an essential key factor to good physiological function [2] and warn about the harmful consequences resulting from abnormalities in the sleep duration [3]. These can be physical and psychological problems, several disease states, and even higher mortality. In this regard, we look to the formulated conclusion given by [2], whereby sleep quantity and sleep quality determine its restorative function and allow us to This research was partially funded by the EU Interreg V-Program "Alpenrhein-Bodensee-Hochrhein": Project "IBH Living Lab Active and Assisted Living", grants ABH40, ABH41 and ABH66. This research was also partially funded by EDITH Research Project (PGC2018-102145-B- C21,C22 (AEI/FEDER, UE). Thomas Penzel was partially supported by the Russian Federation government mega-grant Nº 075-15-2019-1885. M. Gaiduk is with the Ubiquitous Computing Lab at HTWG Konstanz, Alfred-Wachtel-Str. 8, 78462 Konstanz, Germany and with the University of Seville, Avda. Reina Mercedes s/n, Seville, Spain (phone: +49 7531 206-703; Email: [email protected]). ORCID: 0000-0001-9607-5709 J.J. Perea is with the University of Seville, Avda. Reina Mercedes s/n, Seville, Spain (Email: [email protected]). ORCID: 0000-0003-1630-7579 R. Seepold is with the Ubiquitous Computing Lab at HTWG Konstanz, Alfred-Wachtel-Str. 8, 78462 Konstanz, Germany and Institute of Digital maintain a good health state. Several aspects should be quantified to assess the quality of sleep, including identifying different stages, their sequence, and the duration of each of them. A standard procedure allows evaluating each sleep stage. This procedure is called polysomnography (PSG), and it is performed following the guidelines of the American Academy of Sleep Medicine (AASM) [4]. During PSG, electroencephalography (EEG), electrocardiography (ECG), electromyography (EMG), and electrooculography (EOG) signals are recorded continuously during sleep [5]. This set of parameters is established because each of the signals changes during each phase of sleep. Thus, the combination of the obtained results allows classifying the subject's stage at each moment. Although PSG provides an environment for sleep recording and analysis, its implementation is faced with a set of limitations related to: • logistical and economic cost problems due to the requirement of at least 22 analysis connectors for usual implementation [5, 6]; • the high effort of time and personnel for processing and analyzing of data [7]; • non-natural sleep environment in a sleep lab as well as discomfort due to several electrodes, sensors, and cables attached to the subject’s body affecting sleep pattern [8]. Furthermore, there is a limited number of sleep medicine specialists and sleep laboratories [9], which leads to delays in starting the care of patients with sleep disorders. This care is critical because of the large number of undiagnosed sleep disorders resulting in serious problems, even in death [3]. Because non-invasive sleep stage classification is a relevant research topic, many scientific publications can be found on the Medicine at I.M. Sechenov First Moscow State Medical University, Moscow, Russia (Email: [email protected]). ORCID: 0000-0002-5192-2537 N. Martínez Madrid is with the IoT Lab at Reutlingen University, Alteburgstr. 150, 72762 Reutlingen, Germany and the Institute of Digital Medicine at I.M. Sechenov First Moscow State Medical University, Moscow, Russia (Email: [email protected]). ORCID: 0000-0003-1965-9481 T. Penzel is with the Interdisciplinary Sleep Medicine Center, Charité Universitätsmedizin Berlin, Charitéplatz 1, 10117 Berlin, Germany and with the Saratov State University, Russia (Email: [email protected]). ORCID: 0000-0002-4304-0112 M. Glos, is with the Interdisciplinary Sleep Medicine Center, Charit Universittsmedizin Berlin, Charitplatz 1, 10117 Berlin, Germany. (Email: [email protected]). ORCID: 0000-0002-7819-0444. J.A. Ortega is with the University of Seville, Avda. Reina Mercedes s/n, Seville, Spain (Email: [email protected]). ORCID: 0000-0001-9095- 7241 Estimation of Sleep Stages Analyzing Respiratory and Movement Signals Maksym Gaiduk, Juan José Perea Rodríguez, Ralf Seepold, Natividad Martínez Madrid, Thomas Penzel, Martin Glos, and Juan Antonio Ortega S This article has been accepted for publication in IEEE Journal of Biomedical and Health Informatics. This is the author's version which has not been fully edited and content may change prior to final publication. Citation information: DOI 10.1109/JBHI.2021.3099295 © 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See https://www.ieee.org/publications/rights/index.html for more information. Reprinted, with permission, from M.Gaiduk et al. Estimation of Sleep Stages Analyzing Respiratory and Movement Signals, IEEE Journal of Biomedical and Health Informatics, February 2022 M. Gaiduk et al.: Estimation of Sleep Stages Analyzing Respiratory and Movement Signals 1 subject (e.g. [10-16]). However, they use different vital signals as the input for the classification algorithm. A relatively small part of algorithms can work with only movement and breathing signals, as it is done in our approach. We will present several publications relevant to this research case. The main focus will be on research studies conducted using body movement and respiration as vital signals to recognize sleep stages. Scientific papers utilizing other input signals (e.g., heart rate variability (HRV) in [12] or EEG in [15]) are beyond the scope of this literature review. Using body movement data for the identification of sleep stages is proposed in [13]. Sixteen healthy people have participated in the experiment. They slept through the night with electrodes attached to their bodies. NapVIEW with an infrared motion sensor was used as the device to detect body movements, placed about nineteen inches away from the subject's head. With this device's help, the body movement density (BMD) was calculated as the number of body movements per time unit to be used as an index of sleep transitions. The results of this study have confirmed the strong relation of BMD with sleep stage transitions. Besides that, the argument has been made that a BMD cycle is less affected by individual variations, and it is, for this reason, a more accurate index for sleep cycle identification than the absolute value of BMD. Body movement and breathing can be used to classify sleep stages, as presented in [17]. Twenty-five features based on the respiratory signal (depth-based and volume-based) were designed, and the accuracy of 72.3% with Cohen's kappa of 0.34 was obtained for the classification of Wake/NREM/REM stages without subject-specific normalization. If only fourteen best features were used, the accuracy of 71.7% and Cohen's kappa of 0.32 are reported. A considerable correlation between the respiratory signal and sleep stages was demonstrated in [17], a study that constitutes its importance as research on noninvasive sleep phase identification. Moreover, ultra-low-power reflected radiofrequency waves could be used to determine sleep/wake states according to [16]. The proposed algorithm is based on an analysis of movement and, to a lesser extent, respiration signals. The system's evaluation was performed on 113 subjects (94 males, 19 females with an average age of 53±13years). The accuracy of 78% and Cohen's kappa of 0.38 compared to PSG measurement were achieved. One study [14] presents the comparison of wrist and chest actigraphy combined with HRV to identify sleep stages. Even if this approach uses HRV and movement signal, it is relevant for our work because of a movement role in identifying sleep stages. A support vector machine (SVM) was proposed as the approach for the classification of sleep and wake stages. The methodology was tested on a group of 18 healthy adult subjects. The accuracies of about 77% for wrist and 78% for chest actigraphy combined with HRV were achieved. As the respiration signal is analyzed to identify sleep stages, it is essential to know if obstructive sleep apnea (OSA) affects identification accuracy. Otherwise, an algorithm providing accurate results for the patients without OSA could have decreased accuracy for the persons with OSA. This research point was investigated in [18], where the accuracy of 70.9% was achieved, and no significant differences in the identification of sleep stages for non-OSA and OSA groups were noticed. This matter allows transferring the classification results obtained with non-OSA test groups to OSA patients. However, the difference cannot be excluded entirely due to different possible sleep phase identification algorithms. In [11], an approach of identification of Wake/NREM/REM states is presented. The input signals are respiration and movement, and the extracted out of these signals features used for the classification of sleep stages are the following: respiration rate variability, respiration rate, leg movement, body movement, posture, and body orientation features. The signals used for the algorithm's functioning can be obtained in a nonobtrusive way, for example, using the pressure-sensitive etextile bed sheet, as it was done in [11]. The experiment with seven subjects (3 male, 4 female, age: 21-60 years old) was performed to evaluate the approach using three different classifiers: K-nearest Neighbor, Support Vector Machines, and Naïve Bayes. The best classification results were achieved using the Naïve Bayes classifier with the precision of 70.3%, recall of 71.1%, and total accuracy of 72.2%. To overcome the limitations and the problems of a classical PSG approach mentioned above and of alternative presented solutions, the main research aim of this study is defined as a development of techniques for pre-processing and analyzing data in order to detect sleep-related pathologies, with simplified procedures apt even for personnel who are not especially qualified and without the need of expensive laboratory installation. In sum, the diagnostic system under examination is low-cost, and its minimally invasive data collection methods help maintain normal sleep conditions, which in turn increase the accuracy of results. Considering these premises, we have developed a technique to analyze and classify the sleep stages. The main innovations of our proposal are: 1) reduce the directly recorded signals exclusively to breathing and body movement. Thus, we can reduce the invasive action (compared to EEG or ECG measurement) on the subject to a minimum. Furthermore, these signals may be obtained without necessity of involving trained staff; 2) only four parameters, derived from measured signals, to characterize the sleep stages. This aspect represents a quantitative advantage over existing techniques in which up to more than ten parameters are required [5, 10]; 3) develop a new derived parameter, based on the premises of logistic regression that relates two of the primary biological signals: the patient's breathing as well as movement. The Materials and Methods section contains three subsections. In the Statistical Methods subsection, we will explain the quantitative basis to evaluate the parameters characterizing sleep stages. Once we have described our proposal's quantitative scope, we describe our proposed method in the Proposed Model subsection, justifying both the selection This article has been accepted for publication in IEEE Journal of Biomedical and Health Informatics. This is the author's version which has not been fully edited and content may change prior to final publication. Citation information: DOI 10.1109/JBHI.2021.3099295 © 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See https://www.ieee.org/publications/rights/index.html for more information. In reference to IEEE copyrighted material which is used with permission in this thesis, the IEEE does not endorse any of University of Seville's products or services. Internal or personal use of this material is permitted. If interested in reprinting/republishing IEEE copyrighted material for advertising or promotional purposes or for creating new collective works for resale or redistribution, please go to http://www.ieee.org/publications_standards/publications/rights/rights_link.html to learn how to obtain a License from RightsLink. M. Gaiduk et al.: Estimation of Sleep Stages Analyzing Respiratory and Movement Signals 1 of the parameters and the coefficient proposed to optimize the results obtained. In the Implementation subsection, we describe the details of the implemented model. The obtained results are then presented in the Results section and discussed in the Discussion section. Finally, we mention the possible improvement options and plans for future research in the Conclusion and Outlook section. II. METHODOLOGY A. Statistical Methods This section describes the statistical methods used in our sleep analysis model and justifies this study's preference for statistical methodology. Furthermore, a qualitative explanation of the statistical method is carried out, and the equations used in our model are defined. The proposed software system is designed to identify sleep stages out of breathing and movement signals. For that, a set of sensors measuring both these signals (e.g., as described in [19, 20]) can be used to quantify a subject's biosignals. These sensors generate numerically quantifiable electrical impulses, the sample values, which change at each of the sleep stages. However, although the values change at each stage, it is only possible to perform a sleep analysis after numerical processing using statistical methods. After processing, a set of the derived parameters is obtained, with which it is possible to analyze the evolution of sleep. From a statistical point of view, the derived parameters are independent variables, while the sleep stages are identified as dependent variables. A research review was conducted to select the appropriate statistical approach. It showed that several studies had proposed different statistical methodologies to determine the correlation between independent variables (derived parameters) and dependent variables (sleep stages), with regression-based analysis providing the best results compared to other statistical approaches [21, 10]. Notably, in regression-based analysis, the dependent variable is predicted or obtained from a modulated weighting of the available independent variables [22]. Regardless of whether a specific correlation can be established between independent variables, the main objective in the field of regression analysis is to establish a mathematical relationship between dependent and independent variables [22]. Depending on the numerical characteristics of the independent variables, there are different statistical methods to obtain this relationship between dependent and independent variables. In our study, we are going to focus on multinomial logistic regression (MLR). MLR is an extension of the logistic regression with which it is possible to manage independent variables (the derived parameters) in multiple categories. This choice is due to two factors: the first is that the outcome of a multicategory variable (a variable that can achieve a limited number of categories) needs to be predicted as a function of the independent variables; the second is that these independent variables form a random conditional field or set [23]. In the case of sleep studies, a set is related to vital functions such as breathing or movement of the subject under study. Another advantage of MLR compared to many other algorithms is the fact that the output is not presented as just one value with the detected sleep stage but as a set of probabilities for all sleep stages for each epoch. This allows a further processing and adjustment of the results. Quantitatively in MLR, the odds ratio is the logistic regression coefficient's antilogarithm, which simplifies the calculations and allows a better obtaining of the dependent variables. It should be noted that the odds ratio presents information on how the change of independent variable influences the probability of being in a category versus being in the reference group. If an odds ratio value is greater than 1, then as an independent variable in the interval increases, so does the probability value. Otherwise, if the odds ratio is less than 1, then there is an opposite scenario - the probability value decreases as an interval-independent variable increase [24]. To quantify the description made of MLR, we will introduce the equations that we will implement in our study. To carry out this task, we will designate the results of the categorical random dependent variables as Y ∈ {0, 1, 2 ... k}. According to MLR, the conditional mean of the dependent categorical variables is defined as a logistic function, of a related combination, of independent variables, usually called x. The relationship is defined as 𝐸[𝑌|𝑥]= 𝜎(𝑐𝑇𝑥), (1) where c is the vector of unknown coefficients and where the logistic function, σ, satisfies: 𝜎(𝑥)=1 1+𝑒−𝑥. (2) Considering (1) and (2), it is possible by means of MLR to determine the coefficients vector c that maximizes the probability of observations. This operation is defined as ∏𝑃𝑟(𝑌 = 𝑌𝑖|𝑋𝑖) 𝑛 𝑖=1 , (3) which is equal to the maximization of the probability of registration, which is estimated as ∑𝑙𝑜𝑔𝑃𝑟(𝑌 = 𝑌𝑖|𝑋𝑖) 𝑛 𝑖=1 . (4) This expression is used to formulate the simplified function to maximize the probability: 𝑙(𝑐)= −∑𝑙𝑜𝑔(1+𝑒−1𝑦𝑖∗𝑐𝑇𝑥𝑖) 𝑛 𝑖=1 . (5) Additionally, in MLR, the standard error coefficient is of particular importance, which is calculated as 𝑠𝑒(𝑐𝑖) = ((𝑋𝑇𝐴𝑋)−1)𝑖𝑖, (6) where the Hessian of H = -XTAX and A = diag(a1, ..., an) is the diagonal matrix with ai = σ(cTx) * σ(cTx) [25]. Finally, to know the maximum probability, it is necessary to test the predictors' meaning in the logistic regression [23, 10]. For this purpose, the Wald z-test [25] defined by the equation is usually used: 𝑧𝑖=𝑐𝑖 𝑠𝑒(𝑐𝑖). (7) Knowing this last coefficient, it is necessary to demonstrate the predictors' meaning in the logistic regression. The value of the so-called Wald p-value for the coefficient i is used. The probability of obtaining a value at least as extreme as the observed one is determined, for which the null hypothesis has This article has been accepted for publication in IEEE Journal of Biomedical and Health Informatics. This is the author's version which has not been fully edited and content may change prior to final publication. Citation information: DOI 10.1109/JBHI.2021.3099295 © 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See https://www.ieee.org/publications/rights/index.html for more information. M. Gaiduk et al.: Estimation of Sleep Stages Analyzing Respiratory and Movement Signals 1 to be imposed, that is, (ci = 0). The formula for the value of the Wald p-value for the coefficient i is 𝑝𝑖=𝑃𝑟(|𝑍|≥|𝑧𝑖|) = 2∗(1− 𝐹(|𝑧𝑖|)), (8) where F is the cumulative density function of a standard normal distribution [25]. B. Proposed model The characteristics of the proposed model are described here. We will justify the choice of the parameters derived for identifying each of the three sleep stages and how we can drastically reduce the number of these parameters from 10 to 4 compared to the previous research [10]. We will also introduce a new coefficient that allows us to compare two sources of information, breathing and body movement; we will justify the formulation of this coefficient and describe how this new coefficient can improve the accuracy of the results obtained. Several investigations have revealed a relevant fact from the analytical point of view: there is a direct correlation between biosignal patterns, quantifiable using electronic sensors, and the different sleep stages. Thus, from this correlation of sleep stages-biosignal patterns, it is possible to detect and quantify the wakefulness-sleep situation in which a subject of study can be found. To carry out this task, it is necessary to implement recognition algorithms analyzing parameters obtained from biosignals, quantify and detect the different sleep stages. Some of the essential characteristics of biosignals are summarized in Table 1 extracted from [26-29]. TABLE I RELEVANT CHARACTERISTICS OF SLEEP STAGES [26-29] As a result, we have developed a minimally invasive identification approach in which we only need two sources of biosignal data: (a) body movement, (b) respiration. This selection is based on specific changes in physiological features of the human body caused by sleep. When moving from the wake stage to deeper sleep, both a body's movement (amplitude and frequency) and, at the same time, a heart rate are being decreased [26], so only one of these two data sources needs to be used. To achieve our goal of using signals that can be obtained with less effort and more comfort, we have decided to use a movement signal and refrain from using a heart signal. Furthermore, after a transition from NREM to REM stage, body movement is typically decreased even more or absent [27]. The absence of movement in the REM stage also helps distinguish it from the WAKE stage with a high level of accuracy. Additionally, to obtain more accurate results, we use a respiratory signal and have developed a new derived parameter, integrating body movement and breathing. This will be described later and presented in (16). It is known from [26] that respiration becomes more frequent and less rhythmical in the REM stage, leading to changes in respiratory volume and supports the differentiation between NREM and REM stages. To quantify body movement in our model, which builds on the research developed by [10], the average value of the position changes for all three axes will be calculated according to the equations: 𝑋 = 1 𝑛∑ |𝑥𝑖−𝑥𝑖−1| 𝑛 𝑖=1 (9) 𝑌 = 1 𝑛∑ |𝑦𝑖− 𝑦𝑖−1| 𝑛 𝑖=1 (10) 𝑍 = 1 𝑛∑ |𝑧𝑖−𝑧𝑖−1| 𝑛 𝑖=1 , (11) where xi, yi, zi and xi-1, yi-1, zi-1 are the sensor coordinates in the current and correspondingly previous moment. From the (9), (10), (11), we calculate the instantaneous position of the body at the moment i as the modulus of the changes in coordinates, i.e.: 𝐵𝑜𝑑𝑦𝑖= √𝑋2+ 𝑌2+ 𝑍2, (12) where X, Y and Z are the signal values for each axis of a 3D activity sensor used in the study. If the signal provided by the sensors measuring movement is 1-dimensional, it can be directly used as the input for (13). The derived body movement parameter is defined as the mean value of the body movement (BM) signal. It is defined as 𝐵𝑀(𝑘)=1 𝑛∑𝐵𝑜𝑑𝑦𝑖 𝑛−1 𝑖=0 , (13) where n is the number of body movements and Bodyi is described in (12). Additionally, to quantify respiration, we will consider the recommendations given by [17, 10]. In these studies, up to four parameters are formulated from the data provided by the breathing sensors. The innovation of our proposal is that we only need two of these parameters to obtain satisfactory results. Performed tests with different sets of breathing-related parameters have indicated no significant improvement of the classification rate in increasing the number of used derived parameters. Therefore, in our proposal, we will only need • Mean respiratory depth of exhalation (Tsdm) • Median respiratory volume during inhalation (Vin) The choice of these parameters is conditioned by several clinical studies in [17, 26] in which the importance of considering mean respiratory depth, mainly in terms of exhalation, is highlighted, so we have chosen Tsdm, as well as the Median respiratory volume during inhalation, should already reflect the changes of respiratory effort signals and these, in turn, are related to the stages of sleep [26]. Selecting only two parameters instead of the four proposed by [26] does not imply loss of information since the biosignal data are related to each other [30], so a proper selection allows knowing all the information, as we will demonstrate in the Results section. Stage Characteristics Wake Respiration is typically stable and more frequent than in the NREM (especially in Deep Sleep) stage. Typically, more movement than in NREM (especially in Deep Sleep) and REM stages. NREM When sleep deepens from the WAKE stage, body movements become smaller and less frequent. The deeper the sleep, the less frequent the heart rate and respiration. After some instability in respiration during the Light Sleep stage, respiration becomes more stable when sleep deepens. REM Heartbeat and respiration become more frequent and less rhythmical. Body movement is typical for the epochs just before and directly after the REM stage. Anti-gravity muscles lose the tension, and therefore, the movement is typically absent. This article has been accepted for publication in IEEE Journal of Biomedical and Health Informatics. This is the author's version which has not been fully edited and content may change prior to final publication. Citation information: DOI 10.1109/JBHI.2021.3099295 © 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See https://www.ieee.org/publications/rights/index.html for more information. M. Gaiduk et al.: Estimation of Sleep Stages Analyzing Respiratory and Movement Signals 1 Here is the mathematical representation of the derived breathing parameters, 𝑇𝑠𝑑𝑚(𝑘)=𝑚𝑒𝑑𝑖𝑎𝑛(𝑡1,𝑡2,...𝑡𝑛) 𝐼𝑄𝑅(𝑡1,𝑡2,...𝑡𝑛), (14) 𝑉𝑖𝑛 =(𝑘)𝑚𝑒𝑑𝑖𝑎𝑛(∑𝑠𝑥, 𝑠𝑥∈𝛺1 𝑖𝑛 ∑𝑠𝑥,...,∑𝑠𝑥 𝑠𝑥∈𝛺𝑘 𝑖𝑛 𝑠𝑥∈𝛺2 𝑖𝑛 ),(15) where t = t1, t2, ..., tn are the sequences of peaks and troughs of a selected time window, IQR is the interquartile range for the given sequences of peaks, the kth inhalation and exhalation cycle is defined through Ωink with k consecutive breathing cycles (k = 1, 2, ..., K) and Sx is the respiratory effort value. It is important to note that not absolute values of the derived from respiratory signal parameters are significant for the correct functioning of the proposed algorithm, but their changes over time. Furthermore, all signals are being subject-normalized and do not contain raw values. Therefore, even not calibrated sensors (e.g., inductive plethysmography) can be used for signal recording. However, with these three parameters, defined in (13), (14), and (15), it is impossible to obtain precise results that allow the determination of the sleep phase. In order to avoid this problem and achieve a higher precision from the three derived parameters, we formulated a new derived parameter 𝐵𝑉𝑖𝑛 which represents the combination of the inhalation parameters 𝑉𝑖𝑛 and the 𝐵𝑀(𝑘) according to the equation, 𝐵𝑉𝑖𝑛 =𝑙𝑛 𝐵𝑀(𝑘) 𝐵𝑀(𝑘)+𝑉𝑖𝑛. (16) This choice is not arbitrary but is based on two fundamental conclusions: the first is that when the subject under study evolves from the waking stage to that of a deep sleep, the movement of the body decreases, and the depth of breathing increases [17, 26] so that the combination in the form of a coefficient will improve the accuracy of detecting the sleep stage. The amplitudes of the breath and body movement signals can vary significantly between subjects. To standardize these deviations, BM(k) is divided by BM(k) + Vin The second is that using the coefficient as an argument for a logarithmic function reduces the large fluctuations associated with body movement, as highlighted by [26]. Tests were also conducted in which all selected parameters were excluded from the algorithm one by one. This resulted in a reduction in accuracy of about 10-15%, with no clearly detectable differences depending on which parameter was excluded. Thus, by calculating three derived parameters and a fourth obtained from them, we will demonstrate that accurate results are obtained, as presented in the section Results. C. Implementation In our study, we have implemented a classification model based on the statistical technique of MLR, in which we identify as dependent variable (Y) the various stages of sleep, while the independent variables (X1, X2, ..., Xn) are the biosignal parameters which in our case are derived out of breathing and body movements, as described in the section Proposed Model. To calculate the derived parameters, the data obtained with a sampling rate of 32Hz was split into epochs of 30 seconds each. The general structure of the system modules is presented in Figure 1. First, the input signals are pre-processed to facilitate further elaboration, then the derived parameters are calculated, and the calculation of probabilities of sleep phases for every epoch is performed. After each step, the results of the calculations are saved in a separate file to enable a subsequent analysis of the operations carried out and simplify the exchange of system modules if required. The next step is to determine the most probable sleep phase for each epoch and finally to visualize the results. The input signals can be recorded independently or as a part of a polysomnography study. For breathing signals, a chest inductive plethysmography (RIP) record may be used. Other sensors that can record respiratory signals, including changes in their amplitude, can also be applied [e.g. 19, 20]. Body movement signals can be replaced by a signal monitored by a 3D acceleration sensor in a recording device placed on the subject's chest. Another option for the measurement of movement signal could be pressure sensors placed under the mattress [19]. The baseline wandering should be excluded from the lower frequency range and not be significant for the final results, as only the changes in the signal will be analyzed and not its absolute value. Also, the baseline shift has a much smaller amplitude than body movements. A median filter was applied to the respiratory signal to remove short, highamplitude body movements. Afterward, a rolling mean filter was applied to get a cleaner signal, which was beneficial for calculating the selected derived parameters. To increase the accuracy of the classification results, the Fig. 1. System structure. This article has been accepted for publication in IEEE Journal of Biomedical and Health Informatics. This is the author's version which has not been fully edited and content may change prior to final publication. Citation information: DOI 10.1109/JBHI.2021.3099295 © 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See https://www.ieee.org/publications/rights/index.html for more information. M. Gaiduk et al.: Estimation of Sleep Stages Analyzing Respiratory and Movement Signals 1 transition patterns between the different sleep stages were taken into account in implementing the algorithm described in this work. As shown in [31], some transitions between sleep stages are much more likely than others, and at the same time, some transitions have a very low probability. In the first implementation of the algorithm, the emphasis was placed on non-probability transitions. As the algorithm works with probabilities, it allows additional adjustments, increasing accuracy without having to comply with strict rules. In this case, the algorithm reduced the probability of the examined sleep stage by 10-15% (a higher percentage would negatively affect the total classification accuracy if the previous stage were incorrectly classified) if its probability of occurrence according to the transition pattern is almost zero. This approach has led to an increase in the proposed algorithm's accuracy by up to 3% compared to implementation without considering transition probabilities. III. RESULTS The evaluation of the algorithm's work was performed with the use of the dataset provided by the Charité1 clinic in Berlin. The experimental procedures involving human subjects described in this paper were approved by the Institutional Review Board of the Charité-Universitätsmedizin Berlin (application number: EA1/320/114). The overnight recordings of 35 persons (with a total length of about 260 hours) were available for evaluation. The participants' average age was 38.6 +/- 14.5 years old, and the BMI averaged 24.4 +/- 4.9 kg/m². The number of male and female subjects was similar, and no significant health disorders were known for the test persons. The used PSG recordings were previously manually analyzed by sleep medicine physicians, and every 30-second epoch was tagged with the corresponding sleep stage (Wake, N1, N2, N3, REM). As the classification algorithm operates with three stages (Wake, NREM, REM), the stages N1, N2, and N3 from the PSG recordings were merged to the evaluation's NREM stage. The available dataset was strictly separated into training and test subsets. The training dataset consisted of sleep data of randomly selected subjects (after separation male/female) with a total amount of about 100 hours of recordings. Typically, the NREM sleep stage is the prevalent one during sleep, resulting in classification errors due to unbalanced classes in the training dataset. Therefore, it was necessary to balance the subset used for the system training, which was performed during the 1 Initial study was carried out in Charité - Universitätsmedizin Berlin Center of Sleep Medicine Charitéplatz 1, D-10117 Berlin (Germany). evaluation preparation phase. The test dataset included about 160 hours of overnight recordings of 20 subjects, which corresponds to 19324 epochs, 30 seconds each. Both subsets had a similar male/female ratio, and the average BMI of the included subjects did not have a significant discrepancy. Visual representation of sleep stages estimation is presented in Figure 2. It can be seen that there are multiple rapid transitions from one state to another at some points in time. It happens because there is no algorithm for averaging the results of sleep stage classification implemented, and every single epoch is directly visualized according to calculated values. Table 2 represents the classification results of the developed algorithm compared to the expert classification. The rows of the table contain the number of epochs corresponding to each sleep stage according to the expert’s evaluation, whereas the columns represent the results of classification by the developed algorithm. The main diagonal indicates the number of sleep epochs, where the results of both classifications (by experts and algorithm) are in accordance. The achieved general accuracy is equal to 73% following the expert's classification. More statistical measurements are presented in Table 3. In addition to the classification accuracy, we estimated Cohen’s kappa parameter, which is commonly used to measure agreement between several observers (in this case – methods). Its calculated value for the developed classification algorithm compared to the results of experts' classification is 0.44. TABLE II CLASSIFICATION RESULTS Stage Expert Stage developed system Wake NREM REM Total Wake 1256 862 604 2722 NREM 1104 11081 1691 13876 REM 211 730 1785 2726 Total 2571 12673 4080 19324 TABLE III STATISTICAL MEASUREMENTS Overall accuracy 73.0 % Recall Precision Wake 46.1 % 48.9 % NREM 79.9 % 87.4 % REM 65.5 % 43.8 % 1 2 3 1 17 33 49 65 81 97 113 129 145 161 177 193 209 225 241 257 273 289 305 321 337 353 369 385 401 417 433 449 465 481 497 513 529 545 561 577 593 609 625 641 657 673 689 705 721 737 753 769 785 801 817 833 849 865 881 897 913 929 945 Sleep stage (1=Wake, 2=NREM, 3=REM) Epoch Hypnogram Fig. 2. Results of classification of sleep stages for one subject visualized as a Hypnogram. This article has been accepted for publication in IEEE Journal of Biomedical and Health Informatics. This is the author's version which has not been fully edited and content may change prior to final publication. Citation information: DOI 10.1109/JBHI.2021.3099295 © 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See https://www.ieee.org/publications/rights/index.html for more information. M. Gaiduk et al.: Estimation of Sleep Stages Analyzing Respiratory and Movement Signals 1 IV. DISCUSSION This manuscript research aimed to provide a scientific base for the development of a sleep study system that could be used in medical or home environments. Monitoring movement and breathing of recumbent subjects in a non-obtrusive way is less challenging than monitoring heart signals (especially HRV) [32]. It was imperative to develop an algorithm that could classify sleep stages relying on only these two signals for input, the target of the performed research. The achieved general accuracy of algorithm function of 73% indicates that the goal of this challenging task was met. The developed algorithm recognized the NREM stage with high recall of about 80%, which is higher than for both other stages – Wake and REM. The overestimation of the NREM phase can partially explain its prevalence in a typical sleep pattern. Even using a balanced training dataset did not completely solve this problem. Another critical point is that human respiration and movement during the Wake stage just before falling asleep and during the NREM1 stage, which is a part of the NREM stage performed in this work evaluation, are very similar [33]. It may lead to misclassification of the Wake stage as the NREM stage and vice versa. Significant differences in accuracy for different sleep stages are also typical for other systems (e.g. [11, 40]). Another factor that has a strong influence on the results is the quality of the input signals. When analyzing the recognition rate per person, it was detected that there are some recordings where the results are significantly worse than the average. Accordingly, such exceptional cases have driven down the average accuracy and average Cohen's kappa value. After a thorough analysis of such recordings, it was found that the signals had significant differences from other typical recordings. Figure 3 shows a comparison of two recordings of approximately 130 minutes each of the respiratory signal compressed in time. The upper graph in Figure 3 illustrates a typical recording (accuracy of classification 76%) in which the amplitudes of the respiratory signal have few outliers and can therefore be adequately analyzed by the algorithm. However, when looking at the lower graph in Figure 3 (classification accuracy 49%), one can notice that the signal is volatile and contains many disturbances that make it almost impossible to analyze the amplitude of the signal (and also its volume) in numerous signal sections. Nevertheless, the developed algorithm has proven that the accuracy of the classification of sleep stages remains adequate even with the inclusion of such limiting cases. If we would exclude this kind of cases, we could significantly increase the accuracy. However, the presented work aimed to test the functionality of the developed system under natural conditions, which also means the inclusion of low-quality signals. Cohen's kappa value of 0.44 (which is “fair to good” according to [34]) may be considered a good result because a very high value can be challenging to achieve for epoch-by- epoch sleep stage identification. To clarify, even for the AASM scoring standard, the evaluation by different experts has an overall level of agreement of 82% with Cohen's kappa equal to 0.76 [35]. It is, however, important to mention that in [35], five stages were considered. It is expected that a sleep stage identification with only movement and breathing signals as input will have lower levels of accuracy and Cohen's kappa than the gold standard, which uses PSG signal as input. The main novel point proposed in this study's approach is that it uses only the signals that can be obtained in a non-obtrusive way: movement and breathing [19, 20], for which a unique set of parameters was designed. For that purpose, comprehensive literature research and statistical analysis were performed. The new combination of derived parameters and MLR-approach, extended by algorithms for considering transition probabilities from one sleep stage to another, has led to the development of a new unique software solution for identifying sleep stages. A significant amount of research on automatic sleep stage classification is based on using the EEG signal as the algorithm's input [36-38]. As EEG can be recorded only in an obtrusive way with electrodes attached to the subject’s head, these researches are not directly comparable with that presented in this article’s approach. Another large subset of sleep research considers the identification of sleep stages, having the heart rate or ECG signal as the input. In [39], the authors used the ECG signal and respiration effort. The accuracy of 80% for the classes Wake, Fig. 3. Respiration signal. Normal (upper) and faulty (lower) recordings. This article has been accepted for publication in IEEE Journal of Biomedical and Health Informatics. This is the author's version which has not been fully edited and content may change prior to final publication. Citation information: DOI 10.1109/JBHI.2021.3099295 © 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See https://www.ieee.org/publications/rights/index.html for more information. M. Gaiduk et al.: Estimation of Sleep Stages Analyzing Respiratory and Movement Signals 1 NREM, and REM is slightly higher than in our approach. However, it would not be possible to perform the measurement described in [39] in a non-obtrusive way because the ECG signal is being measured directly by placing electrodes on the human body. Therefore, our method has a significant advantage as the necessary signals may be measured in a contactless way as described in [19, 20], while reporting only a slight reduction of accuracy. Another crucial advantage of our approach is using only four features, compared to 80 in [39], which means faster processing and less computing power needed, which is essential for the embedded systems. To continue, the approach described in [40] uses the HRV signal derived from the ECG. The achieved accuracy of 72,6% for subject independent classifier is slightly lower than the accuracy of the method presented in our paper. Though the accuracy is comparable, our method is preferable since it uses only respiration and movement signals (which can be obtained contact-free). Furthermore, the algorithm's number of features used is significantly lower than proposed in [40] (4 against 41). Breathing and body movement signals are used as the input in [16] approach, making it similar to our method concerning analyzed signals. However, only Wake/Sleep states are recognized in [16], whereas our system identifies Wake, NREM, REM sleep stages. The accuracy of 78%, with a Cohen’s kappa of 0.38 stated in [16], compared to 73% of accuracy with a Cohen's kappa of 0.44 in our experiment, confirms the quality of our results. The number of recognized stages in our method is higher, albeit with a marginal decrease in accuracy, and it achieved a better Cohen's kappa value. The reported results in [17] regarding the accuracy of 72.3% with Cohen's kappa of 0.34 (25 features), and accuracy of 71.7% with Cohen's kappa of 0.32 (14 features) for the classification of Wake/ NREM/REM stages without subjectspecific normalization is lower than in our work. Our work also presents data using fewer used features (4 against 25/14), another advantage over other studies. The respiratory signal is used in [17] as the input for the algorithm. Compared to the approach presented in [11], our method uses the same input signals but indicates better results: 73% accuracy against 72%, using only four features compared to more than 30 in [11]. Furthermore, a significantly higher number of overnight recordings was used to evaluate our experiment (20 against 7), which illustrates the results' stability. It is important to note that a reduction in the number of the used features is especially relevant when it does not have any negative impact on the accuracy of detection because the effort required to calculate the features is moderate. The results of the comparison with state-of-the-art solutions presented above confirm that our system can use a small number of parameters and demonstrate high accuracy at the same time. Although the results of the work carried out are promising, some limitations should be addressed: • The number of recordings for the training was limited, and therefore it cannot be excluded that with a significantly larger quantity, the training results could be different. • Changes in the signal in the process of recording (e.g. due to the displacement of the RIP belts) can lead to a reduction in accuracy. • The available recordings were divided into 30-second intervals, which is also typical for sleep analysis. However, transitions from one sleep stage to another can also occur within these intervals, which was not considered in the current implementation of the algorithm. • Although the possibility of substituting the devices for recording the movement and respiration signals is preplanned, it was not possible to use and compare several alternative devices for the recording within the scope of the study conducted. Therefore, it cannot currently be claimed that the results of the algorithm's work will be identical for different signal sources. V. CONCLUSION AND OUTLOOK The development of an algorithm for identifying Wake/NREM/REM sleep stages was performed, and the accuracy of 73% with the Cohen's kappa value of 0.44 was obtained. These results confirm the proposed approach's appropriateness for the defined use case with only breathing and movement signals as the input. Further investigation on this method promises the algorithm's improvement and, consequently, more accurate results of system work. One possible way to enhance the algorithm's performance is to continue investigating the best set of used derived parameters and the development of new ones. However, this needs an indepth investigation of the importance of different features and values in the general sleep stages recognition algorithm. This work was done, but further improvements should still be possible and are planned for future projects. Moreover, other sleep stages recognition approaches could be combined with the one proposed in this study to get an extended algorithm, which would obtain the final identification results as a mix of two or more methods, e.g., MLR + neural network. In this case, the epochs with the same identified stages by both approaches could be used as "trust-anchors," and the epochs before and after them could be identified more precisely, considering the transitions between the different sleep states [28]. Having three or more combined approaches, the decision system could be implemented, determining the final sleep stage identification result as the election by the majority of algorithms. As shown in the Discussion section, the quality of the input signal plays an essential role in the correct classification of sleep stages. Therefore, in the future version of the system, it is planned to label the recordings with the disturbed signals and, if necessary, to remove some parts of the signal from the analysis and to provide a corresponding notification if no signal correction should be possible. Signal reconstruction for improving the quality of the signal is a topic in itself and could also be included in future research. This would make it possible to perform a good detection of sleep phases even in parts of the signal where no reliable evaluation was initially possible. Another possible direction of future work is to develop a standalone system tuned to a home environment [41], which requires developing a hardware part of the system to complement the software. This development's significant progress is already made and presented in several scientific publications [19, 42]. The final aim is to gain a system that This article has been accepted for publication in IEEE Journal of Biomedical and Health Informatics. This is the author's version which has not been fully edited and content may change prior to final publication. Citation information: DOI 10.1109/JBHI.2021.3099295 © 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See https://www.ieee.org/publications/rights/index.html for more information. M. Gaiduk et al.: Estimation of Sleep Stages Analyzing Respiratory and Movement Signals 1 would recognize sleep stages and sleep apnea, available for use in a home environment without high financial and personnel costs. Furthermore, the system should be comfortable to use, which has led to selecting the signals that can be obtained in a non-obtrusive way – movement and breathing. This kind of system could be widely used to provide the necessary information to medical professionals, thus enabling early detection of sleep problems and ultimately improving the population's quality of sleep. ACKNOWLEDGMENT We thank the Interdisciplinary Center for Sleep Medicine of Charité-Universitätsmedizin Berlin and in particular Beate Diecker and Claudia Biro for supporting the study. REFERENCES [1] K. K. Gulia and V. M. Kumar, “Importance of Sleep for Health and Wellbeing Amidst COVID-19 Pandemic,” Sleep and vigilance, pp. 1– 2, 2020, doi: 10.1007/s41782-020-00087-4. [2] National Heart, Lung, and Blood Institute, Your guide to healthy sleep. Washington D.C.: U.S. Dept. of Health and Human Services National Institutes of Health National Heart Lung and Blood Institute, 2011. [3] T. L. Lee-Chiong, C. Mattice, and R. Brooks, Fundamentals of Sleep Technology. Philadelphia: Wolters Kluwer, 2015. [4] R. B. Berry et al., The AASM manual for the scoring of sleep and associated events: Rules, terminology and technical specifications: V. 2.6. Darien, Illinois: American Academy of Sleep Medicine, 2020. [5] S. Chokroverty, R. J. Thomas, and M. Bhatt, Atlas of Sleep Medicine, 2nd ed. London: Elsevier Health Sciences, 2013. [6] R. B. Berry, “Indications for Polysomnography, Portable Monitoring, and Actigraphy,” Sleep Medicine Pearls (Third Edition), 2015. [7] M. Hirshkowitz, “Polysomnography Challenges,” Sleep medicine clinics, vol. 11, no. 4, 2016, doi: 10.1016/j.jsmc.2016.07.002. [8] O. Le Bon et al., “The first-night effect may last more than one night,” Journal of Psychiatric Research, vol. 35, no. 3, pp. 165–172, 2001, doi: 10.1016/S0022-3956(01)00019-X. [9] J. Singh et al., “American Academy of Sleep Medicine (AASM) Position Paper for the Use of Telemedicine for the Diagnosis and Treatment of Sleep Disorders,” Journ of clinical sleep med: official publication of the AASM, vol.11, no.10, 2015, doi: 10.5664/jcsm.5098. [10] M. Gaiduk et al., “Automatic sleep stages classification using respiratory, heart rate and movement signals,” Physiological meas, vol. 39, no. 12, p. 124008, 2018, doi: 10.1088/1361-6579/aaf5d4. [11] L. Samy et al., “Unobtrusive Sleep Stage Identification Using a Pressure-Sensitive Bed Sheet,” IEEE Sensors J., vol. 14, no. 7, pp. 2092–2101, 2014, doi: 10.1109/JSEN.2013.2293917. [12] M. Radha et al., “Sleep stage classification from heart-rate variability using long short-term memory neural networks,” Scientific reports, vol. 9, no. 1, p. 14149, 2019, doi: 10.1038/s41598-019-49703-y. [13] Y. Kambayashi and H. Hagiwara, “Estimating sleep cycle using body movement density,” in 2012 5th International Conf. on BioMedical Engin and Informatics, Chongqing, China, Oct. 2012, pp. 1081–1085. [14] M. Aktaruzzaman et al., “Performance comparison between wrist and chest actigraphy in combination with heart rate variability for sleep classification,” Computers in biology and medicine, vol. 89, pp. 212– 221, 2017, doi: 10.1016/j.compbiomed.2017.08.006. [15] N. Michielli, U. R. Acharya, and F. Molinari, “Cascaded LSTM recurrent neural network for automated sleep stage classification using single-channel EEG signals,” Computers in biology and medicine, vol. 106, pp. 71–81, 2019, doi: 10.1016/j.compbiomed.2019.01.013. [16] P. de Chazal et al., “Sleep/wake measurement using a non-contact biomotion sensor,” Journal of sleep research, vol. 20, no. 2, pp. 356– 366, 2011, doi: 10.1111/j.1365-2869.2010.00876.x. [17] X. Long et al., “Analyzing respiratory effort amplitude for automated sleep stage classification,” Biomedical Signal Processing and Control, vol. 14, pp. 197–205, 2014, doi: 10.1016/j.bspc.2014.08.001. [18] S. H. Hwang et al., “Unconstrained Sleep Stage Estimation Based on Respiratory Dynamics and Body Movement,” Methods of information in medicine, vol. 55, no. 6, 2016, doi: 10.3414/ME15-01-0140. [19] M. Gaiduk, R. Seepold, N. Martínez Madrid, S. Orcioni, and M. Conti, “Recognizing Breathing Rate and Movement While Sleeping in Home Environment,” in Lecture Notes in Electrical Engineering, Cham: Springer International Publishing, 2020, pp. 333–339. [20] M. Gaiduk, D. Wehrle, R. Seepold, and J. A. Ortega, “Non-obtrusive system for overnight respiration and heartbeat tracking,” Procedia Computer Science, vol. 176, pp. 2746–2755, 2020, doi: 10.1016/j.procs.2020.09.282. [21] D. C. Montgomery, E. A. Peck, and G. G. Vining, Introduction to Linear Regression Analysis, 5th ed. New York, NY: Wiley, J, 2011. [22] A. von Eye and C. Schuster, Regression analysis for social sciences. San Diego, Calif: Academic Press, 1998. [23] D. W. Hosmer, S. Lemeshow, and R. X. Sturdivant, Applied logistic regression. Hoboken, NJ: Wiley, 2013. [24] M. H. Katz, Multivariable analysis: A practical guide for clinicians and public health researchers, 3rd ed.: Cambridge Univ. Press, 2011. [25] Garson G. D., Logistic Regression: Binary & Multinomial: 2016 Edition: Statistical Associates Publishers, 2016. [26] Y. Kurihara and K. Watanabe, “Sleep-Stage Decision Algorithm by Using Heartbeat and Body-Movement Signals,” IEEE Trans. Syst., Man, Cybern. A, vol. 42, no. 6, pp. 1450–1459, 2012, doi: 10.1109/TSMCA.2012.2192264. [27] A. Sadeh, “The role and validity of actigraphy in sleep medicine: an update,” Sleep medicine reviews, vol. 15, no. 4, pp. 259–267, 2011, doi: 10.1016/j.smrv.2010.10.001. [28] H. C. Dringenberg, Handbook of sleep research. London: Elsevier Academic Press, 2019. [29] N. J. Douglas et al., “Respiration during sleep in normal man,” Thorax, vol. 37, no. 11, pp. 840–844, 1982, doi: 10.1136/thx.37.11.840. [30] D. A. Belsley, Conditioning diagnostics: Collinearity and weak data in regression. New York: Wiley, 1991. [31] A. Schlemmer et al., “Changes of sleep-stage transitions due to ageing and sleep disorder,” Philosophical transactions. Series A, Mathematical, physical, and engineering sciences, vol. 373, no. 2034, 2015, doi: 10.1098/rsta.2014.0093. [32] I. Sadek, J. Biswas, and B. Abdulrazak, “Ballistocardiogram signal processing: a review,” Health information science and systems, vol. 7, no. 1, p. 10, 2019, doi: 10.1007/s13755-019-0071-7. [33] R. D. Ogilvie, R. T. Wilkinson, and S. Allison, “The detection of sleep onset: behavioral, physiological, and subjective convergence,” Sleep, vol. 12, no. 5, pp. 458–474, 1989, doi: 10.1093/sleep/12.5.458. [34] J. L. Fleiss et al., Statistical Methods for Rates and Proportions. Hoboken, NJ, USA: John Wiley & Sons, Inc, 2003. [35] H. Danker-Hopfe et al., “Interrater reliability for sleep scoring according to the Rechtschaffen & Kales and the new AASM standard,” Journal of sleep research, vol. 18, no. 1, pp. 74–84, 2009, doi: 10.1111/j.1365-2869.2008.00700.x. [36] X. Li et al., “HyCLASSS: A Hybrid Classifier for Automatic Sleep Stage Scoring,” IEEE journal of biomedical and health informatics, vol. 22, no. 2, pp. 375–385, 2018, doi: 10.1109/JBHI.2017.2668993. [37] H. Dong et al., “Mixed Neural Network Approach for Temporal Sleep Stage Classification,” IEEE transactions on neural systems and rehabilitation engineering: a publication of the IEEE Engineering in Medicine and Biology Society, vol. 26, no. 2, pp. 324–333, 2018, doi: 10.1109/TNSRE.2017.2733220. [38] M. Diykh, Y. Li, and S. Abdulla, “EEG sleep stages identification based on weighted undirected complex networks,” Computer methods and programs in biomedicine, vol. 184, p. 105116, 2020, doi: 10.1016/j.cmpb.2019.105116. [39] P. Fonseca, X. Long, M. Radha, R. Haakma, R. M. Aarts, and J. Rolink, “Sleep stage classification with ECG and respiratory effort,” Physiological measurement, vol. 36, no. 10, pp. 2027–2040, 2015, doi: 10.1088/0967-3334/36/10/2027. [40] M. Xiao, H. Yan, J. Song, Y. Yang, and X. Yang, “Sleep stages classification based on heart rate variability and random forest,” Biomedical Signal Processing and Control, vol. 8, no. 6, pp. 624–633, 2013, doi: 10.1016/j.bspc.2013.06.001. [41] M. Conti, S. Orcioni, N. M. Madrid, M. Gaiduk, and R. Seepold, “A Review of Health Monitoring Systems Using Sensors on Bed or Cushion,” in Lecture Notes in Computer Science, Cham: Springer International Publishing, 2018, pp. 347–358. [42] M. Gaiduk et al., “Embedded system for non-obtrusive sleep apnea detection,” IEEE Engineering in Medicine and Biology Society. Annual International Conference, vol. 2020, pp. 2776–2779, 2020, doi: 10.1109/EMBC44109.2020.9176075. This article has been accepted for publication in IEEE Journal of Biomedical and Health Informatics. This is the author's version which has not been fully edited and content may change prior to final publication. Citation information: DOI 10.1109/JBHI.2021.3099295 © 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See https://www.ieee.org/publications/rights/index.html for more information. ScienceDirect Available online at www.sciencedirect.com Procedia Computer Science 176 (2020) 1–9 1877-0509 © 2020 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0) Peer-review under responsibility of the scientific committee of the KES International. 10.1016/j.procs.2020.08.001 24 th KES International Conference on Knowledge-Based and Intelligent Information & Engineering Systems KES2020 M. Cristani, C. Toro, C. Zanni-Merk, R. J. Howlett and L. C. Jain (Eds.) M. Cristani et al. / Procedia Computer Science 176 (2020) 1–9 3 Preface This edition of Procedia Computer Science represents the proceedings of the 24th International Conference on Knowledge - Based and Intelligent Information & Engineering Systems (KES 2020), organised by KES International and held in virtual form on the KESonline virtual conference platform, 16-18 September 2020. This is the first time that this conference has been held virtually, and this has been made necessary because of the Covid-19 pandemic which has swept the world. Covid-19 has caused disruption, economic problems, and human tragedy in many countries. We offer condolences and sympathy to those who have sadly lost loved ones or colleagues in the pandemic. KES 2020 was the 24th event in a series of broad-spectrum intelligent systems conferences first held in Adelaide, Australia in 1997. The main aim of this KES conference series is to provide an internationally respected forum for the dissemination of research results and the discussion of issues relating to the theory, technologies and applications of intelligent engineering and information systems. This truly international conference attracted submissions from a substantial number of researchers and practitioners from all over the world, who submitted their papers to four general tracks and 40 special sessions on specific topics. A large number of submissions was received and each paper was peer reviewed by at least two members of the International Program Committee. From them, 401 high - quality papers were accepted for oral presentation and publication in Procedia Computer Science, submitted for indexing in Conference Proceedings Citation Index (CPCI), Scopus and INSPEC. The conference chairs would like to express their gratitude to the Keynote Speakers: Prof Yukio Ohsawa, University of Tokyo, Japan, title of talk: Beyond Explainable Learning to Explanatory Intelligence- A Systems Innovation Approach on Data Jackets -'; Dr Urszula Stańczyk, Silesian University of Technology, Poland, title of talk: On the uniqueness of writing styles- Problems and tasks of stylometry'; Dr. Ralf Christian Haerting, University of Aalen, Germany, ‘title of talk to follow.'; and Dr Guido Governatori, Data 61, Dutton Park, Australia, title of talk: ‘title of talk to follow'. We would like to acknowledge also the Programme Co-Chairs, the General Track Chairs, the International Programme Committee members and reviewers for their valuable efforts in the review process, helping us to guarantee the highest quality possible for the conference. We would also like to thank the organisers and chairs of the special sessions which make an essential contribution to the success of the conference. Lastly, we would like to thank all the authors, presenters and delegates for their valuable contribution in making this an extraordinary event. KES International hopes and intends that KES2020 will make a significant contribution to international research collaboration and understanding, an essential task for the promotion of scientific joint work and excellence. Prof. Robert J. Howlett, ‘Aurel Vlaicu’ University of Arad, Romania and Bournemouth University, UK and Prof. Lakhmi C. Jain, University of Technology Sydney, Australia and University of Canberra, Australia Organisation Honorary Chairs Prof. Lakhmi C. Jain, University of Technology Sydney, Australia and Liverpool Hope University, UK Prof. Vincenzo Manca, University of Verona, Italy General Chair Dr. Matteo Cristani, University of Verona, Italy Publicity Chair Dr. Francesco Olivieri, CSIRO, Data61, Australia Executive Chair Prof. Robert J. Howlett, Bournemouth University, UK Programme Chairs General Track Programme Chair Dr Carlos Toro, A*STAR, Singapore Invited Sessions Programme Chair Prof. Cecilia Zanni-Merk, INSA Normande University, France KES Conference Series KES2020 is part of the KES Conference Series Conference Series Chairs: L. C. Jain and R. J. Howlett ScienceDirect Available online at www.sciencedirect.com Procedia Computer Science 176 (2020) 2341–2349 1877-0509 © 2020 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0) Peer-review under responsibility of the scientific committee of the KES International. 10.1016/j.procs.2020.09.297 © 2020 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0) Peer-review under responsibility of the scientific committee of the KES International. Keywords: Sleep study; sleep quality; PSQI; sleep efficiency 1. Introduction Sleep is an essential part of our life that, on average, takes up to one-third of every day [1]. Taking care of it is crucial for our health, as sleep is necessary for the recovery of our body as well as the brain. The two main components of sleep are duration and quality. Furthermore, sleep quality can also be seen as a combination of several characteristics ∗Corresponding author. Tel.: +49-7531-206-703. E-mail address: [email protected] 24th International Conference on Knowledge-Based and Intelligent Information & Engineering Systems Comparison of sleep characteristics measurements: a case study with a population aged 65 and above Maksym Gaiduka,b,∗, Ralf Seepolda,c, Juan A. Ortegab, Natividad Mart´ ınez Madridc,d aHTWG Konstanz, Alfred-Wachtel-Str. 8, 78462 Konstanz, Germany bUniversity of Seville, Avda. Reina Mercedes s/n, Seville, Spain cI.M. Sechenov First Moscow State Medical University, 2-4, Bolshaya Pirogovskaya st., 119435 Moscow, Russian Federation dReutlingen University, Alteburgstr. 150, 72762 Reutlingen, Germany Abstract Good sleep is crucial for a healthy life of every person. Unfortunately, its quality often decreases with aging. A common approach to measuring the sleep characteristics is based on interviews with the subjects or letting them fill in a daily questionnaire and afterward evaluating the obtained data. However, this method has time and personal costs for the interviewer and evaluator of responses. Therefore, it would be important to execute the collection and evaluation of sleep characteristics automatically. To do that, it is necessary to investigate the level of agreement between measurements performed in a traditional way using questionnaires and measurements obtained using electronic monitoring devices. The study presented in this manuscript performs this investigation, comparing such sleep characteristics as “time going to bed”, “total time in bed”, “total sleep time” and “sleep efficiency”. A total number of 106 night records of elderly persons (aged 65+) were analyzed. The results achieved so far reveal the fact that the degree of agreement between the two measurement methods varies substantially for different characteristics, from 31 minutes of mean difference for “time going to bed” to 77 minutes for “total sleep time”. For this reason, a direct exchange of objective and subjective measuring methods is currently not possible. 2342 Maksym Gaiduk et al. / Procedia Computer Science 176 (2020) 2341–2349 like the length of time to fall asleep, a number of awakenings at night, etc. [2]. It is known that sleep behavior changes with the age [3]. Among other things, the structure of the sleep stages, sleep efficiency, and sleep duration are typically different for elderly persons compared to younger ones. In general, methods for measuring sleep quality can be divided into two large groups - subjective and objective measurement [4]. In objective measurement, the values are recorded and analyzed with electronic devices. Different types of sensors and methods for calculating the sleep quality can be used, such as a combination of a triaxial accelerometer and a pressure sensor [5]. In this work, a system was developed that can detect heart rate, breathing, REM/non-REM sleep phases, movement, the position of the person, sleep apnea, and sleep time, and from these data, it calculates the sleep quality with a self-developed algorithm. However, this system uses an accelerometer that has to be placed on the body. This can interfere with normal sleep patterns. The validation of the results by reference measurements was not clearly described in the article. The recognition of breathing rate and movement while sleeping is also one relevant method [6]. For objective measurement of sleep quality, the sensors on the smartphone can also be used as presented in [7]. In this case, three types of smartphone sensors were used: microphone, accelerometer, and light sensor. The smartphone was also used for keeping the sleep diary (derived from Pittsburgh Sleep Quality Index - PSQI [8]). It achieved a 81.48% of accuracy in the classification of “poor” and ”good” sleepers. However, using the smartphone to measure sleep quality has some disadvantages, such as the need to place the phone close to the person in the bedroom and increased battery consumption during the night. Moreover, the sensors in different smartphones are not the same, which means that the results may differ between them. Subjective measurement is based on recording the perceived sleep quality of the subjects. Questionnaires and/or sleep diaries are used [9]. Electronic sleep diaries are often developed as mobile apps. [10] compares a paper-based sleep diary versus an electronic version, and versus a measurement with an actigraph was made. It was concluded that electronic sleep diaries could have an advantage over classic paper-based versions. Another digital sleep diary was presented in [11]. Users perceived the proposed system as practical, but it was found that user experience and motivational effects should be considered when developing the system. In [12] the results of sleep quality measurements were compared with objective and subjective methods, and a correlation was confirmed. However, such important characteristics as “total tome in bed” or “sleep efficiency” were not considered, and the results of some test persons differ significantly so that a combination of the two methods can be useful. For the detection of sleep disorders (especially insomnias), it is important to pay attention to such sleep characteristics as sleep efficiency, which is also directly linked with sleep duration [13]. Sleep efficiency is defined as the ratio between total time in sleep (sleep time) and the total time in bed (bedtime). It is known that insomnias are more common and sometimes get more distinct in the elderly [14]. Sleep efficiency can be measured objectively (based on electronic device recording) and subjectively (based on the feeling and memory of the person). For the recording of subjective feelings of bedtime and sleep time, the PSQI questionnaire is commonly used [8]. For the electronic measurement, different types of systems can be used, for example, based on the measurement of heart rate, breathing, and movement signals, as described in [15]. In this work, a system based on a multinomial logistical regression approach was developed and tested. The overall achieved accuracy rate is 84% for Wake/Sleep stages, with Cohen’s kappa value 0.46. Several studies targeting the improvement of sleep quality in older adults can be found (e.g., [16,17]). However, to identify the persons with sleep quality problems and to monitor the progress of therapy, an easy-to-use method for broad application is necessary. Therefore, automatically obtaining the results of measurements, making it possible to test essential sleep characteristics for a high number of older adults could support the enhancement of sleep quality for this age group. However, the results of measurements with the help of devices and based on the subjective feeling of a person may have some discrepancies [18], which should be investigated to check the possibility of substitution of questionnaires by automatic measurement with the help of electronic devices. Investigation on this question for the subjects from the age group 65+is the main aim of the study presented in this manuscript. Maksym Gaiduk et al. / Procedia Computer Science 176 (2020) 2341–2349 2343 2. Methodology 2.1. Test subjects A total of 10 elderly persons (five males and five females) has participated in the study. Nine of them were living in their flats and one in a shared apartment for the elderly. All subjects were over 65 years old with mean value (MV) equal to 72.5 and a standard deviation (SD) of 6.2. The average height of participants was 167.6cm (SD=7.9), and the average weight was equal to 80.8kg (SD=12.6). They were able to carry out the majority of the homework themselves and did not have any severe acute diseases. All participants could understand and respond to the interview questions and were not suffering from advanced dementia. They were given all the study information in advance, and participation was voluntary. 2.2. Measurement methods For the measurement of sleep characteristics, two approaches were implemented: (1) questionnaire for the measurement of subjective feeling of participating persons, and (2) electronic device for the measurement of sleep characteristics in an objective way. For the questionnaire-based approach, Pittsburgh Sleep Quality Index (PSQI) was selected [8]. The main reason is that the PSQI is presently the only standardized clinical instrument that covers a broad range of indicators relevant to sleep quality with strong positive evidence for reliability and validity (hypothesis testing) [19]. [20] used the German version of the questionnaire. It asks retrospectively for two weeks about the frequency of sleep-disturbing events, the estimation of sleep quality, regular sleep times, sleep latency and sleep duration, sleep medication intake, and daytime sleepiness. A total of 19 items are used for quantitative evaluation. They are assigned to 7 components, each of which can assume a value range from 0 to 3. The total score results from the summation of the component scores and can vary from 0 to 21, whereby a higher value corresponds to reduced sleep quality. There is an empirically determined cut-offvalue (of 5), which allows classification into “good” and “poor” sleepers. Currently, different devices for sleep tracking can be found. Some of them are based on actigraphy [21]; others use piezoelectric sensors [22] or execute analysis of the acoustic signals [23]. For the study described in this document, the Emfit QS+1sensor, using a ballistocardiography approach, was chosen [24]. This device can be used for sleep tracking, monitoring general characteristics, like time going to bed, total sleep time, time in bed, and light/deep/REM sleep[25]. It is a contact-free sensor, which can be placed under the mattress to ensure the typical sleep pattern of the users without disturbing their sleep. Furthermore, this device works fully-automatic and does not require any action by the participant of the study other than plugging into a power outlet. This was an essential point for the selection, as one of the study goals is to check if it is possible to execute the measurement of sleep characteristics in an automatic way with a minimal number of necessary actions by users and medical staff. All the data collected with the electronic device can only be accessed by the study organizers. 2.3. Study design Participants were informed in advance about the procedure of the study. On the first day, the study organizers installed electronic devices in the participants’ homes. The Emfit device was placed beneath the mattress across the bed under the chest area according to the device manual. Interviews with general questions (like age, sex, height, weight) were also conducted on the first day of study. The test persons were asked to continue with their regular daily routine and to contact the organizers only in case of troubles. The devices were kept by participants for 14 days, during which time they were visited every 3-4 days to check if everything was going well or if there were problems or questions. On the 15th day, the devices were collected by the study organizers, and the PSQI questionnaires were filled out together with the participants. The measurements were executed for 14 nights, and the questions of subjective measurement were also referred to the same 14 days. 1https://www.emfit.com 2344 Maksym Gaiduk et al. / Procedia Computer Science 176 (2020) 2341–2349 2.4. Data evaluation All completed questionnaires were manually processed, and all data were entered into a database. The following sleep characteristics were obtained directly from the questionnaire: time going to bed, sleep onset latency, getting up time as well as total sleep time. PSQI index, total time in bed, and sleep efficiency were calculated according to the guidelines of the German Society for Sleep Research and Sleep Medicine. Calculation of total time in bed: Total time in bed =getting up time −time going to bed (1) Sleep efficiency (quotient of sleep time and bedtime) was calculated as follows: Sleep efficiency =Total sleep time in hours Total time in bed in hours ×100% (2) The sleep records created by the electronic device were downloaded using the developer web interface. Time going to bed, total sleep time, and total time in bed is directly available in the overnight recordings. Sleep efficiency was calculated according to Equation 2. In order to obtain results for a period of 14 nights, which are comparable to PSQI, the average values for the mentioned above sleep characteristics (time going to bed, total sleep, total time in bed, and sleep efficiency) for all 14 nights were calculated. 3. Results and Discussion As several essential sleep characteristics have been measured during the study, the results section will also be divided into six subsections for the detailed presenting of the results. First, the results of two used types of measurements will be introduced: PSQI measurement and measurement with an electronic device (Emfit). After that, several relevant sleep characteristics measured with both measurement approaches will be presented. 3.1. PSQI measurement Table 1presents the PSQI measurement results for 10 subjects. The most relevant characteristics for the comparison with a measurement executed by hardware devices are “Time going to bed”, “Total sleep time”, “Total time in bed” and “Sleep efficiency”. PSQI index is used to distiguish poor and good sleepers [8]. According to this approach, subjects with PSQI index higher than five are identified as poor sleepers. Hence, 6 out of the 10 persons who participated in the study have poor sleep quality. Table 1. PSQI questionnaire results. Subject 1 2 3 4 5 6 7 8 9 10 PSQI Index 5 4 11 6 3 1 6 6 9 7 Time going to bed 22:00 22:45 21:00 00:00 21:00 22:00 21:00 20:00 00:30 00:00 Sleep onset latency (SOL) 00:15 00:30 01:30 00:10 00:10 00:30 01:00 01:00 01:20 00:10 Getting up time 07:30 07:30 06:30 06:00 07:45 06:30 08:00 09:00 08:00 06:30 Total sleep time (TST) 8:00 7:00 6:00 5:30 9:00 8:00 7:00 8:00 5:30 4:30 Total in bed time (TIB) 9:30 8:45 9:30 6:00 10:45 8:30 11:00 13:00 7:30 6:30 Sleep efficiency (SE) 84% 80% 63% 92% 84% 94% 64% 62% 73% 69% 3.2. Electronic device measurement The results of monitoring with the Emfit device, used for the objective measurement of sleep characteristics, are shown in Table 2. The comparison of measurement results of this approach with the PSQI questionnaire is presented in the following sections. Maksym Gaiduk et al. / Procedia Computer Science 176 (2020) 2341–2349 2345 Table 2. Electronic device measurement results. Subject 1 2 3 4 5 6 7 8 9 10 Time going to bed 22:25 22:26 21:13 00:28 20:48 21:22 20:55 20:56 01:18 01:06 Total sleep time (TST) 06:37 08:34 07:50 04:57 09:37 08:30 10:16 10:22 05:20 05:06 Total in bed time (TIB) 07:13 09:37 09:09 05:40 10:41 09:31 11:24 11:21 06:25 05:56 Sleep efficiency (SE) 92% 89% 86% 88% 90% 89% 90% 91% 83% 86% Fig. 1. Average time going to bed. 3.3. Time going to bed The time going to bed for the ten subjects measured with the PSQI questionnaire and electronic device is presented in Fig. 1. Test persons could indicate the going to bedtimes with very high accuracy, which was confirmed by objective measurement results. The average deviation of the time going to bed between PSQI and Emfit measurements is 31 minutes. 3.4. Total time in bed The total time in bed was also estimated very well by the majority of test persons. This can be appreciated in Fig. 2. One of the possible reasons for the bad estimation by subjects 1 and 8 could be that for these two subjects (as well for subjects 2 and 7), records of an electronic device are available for less than eight days of study. Therefore, the calculated average time in bed could differ from the real situation. The average deviation of this parameter for both measurement approaches is equal to 51 minutes for all subjects and 35 minutes in case subjects 1 and 8 are excluded. 3.5. Total sleep time Only 50% of test subjects also have a reasonably good estimation of total sleep time. Fig. 3presents the visualization of this sleep characteristic measurement. These circumstances were also expected because estimating the time spent sleeping is naturally more difficult than total time in bed or going to bed. The availability of a maximum of 2346 Maksym Gaiduk et al. / Procedia Computer Science 176 (2020) 2341–2349 Fig. 2. Average total time in bed (TIB). Fig. 3. Average total sleep time (TST). 7-night records (and not all 14) for subjects 1, 2, 7, 8 may have affected the outcome. Nevertheless, the tendency of underestimating the sleep time can be recognized. This fact is also known from other studies [26]. In the case of the total sleep time measurement, the average deviation between objective and subjective measurement is higher than for the previous two sleep characteristics. It is equal to 77 minutes for all subjects, but it is 54 minutes excluding the subjects 7 and 8 from the statistics, because of a lower amount of available Emfit records for these persons. Maksym Gaiduk et al. / Procedia Computer Science 176 (2020) 2341–2349 2347 Fig. 4. Average sleep efficiency (SE). 3.6. Sleep efficiency The presented in Fig. 4graph of sleep efficiency measurement for both approaches indicates a clear tendency of underestimating this characteristic by the test persons. However, the results of measurement with an electronic device can also be affected by the behavior of test persons – e.g., a continuing lying in bed with no or just a little movement (which can be typical for some persons, especially for less active elderly persons) could be wrongly recognized as sleep by the utilized device. Comparing the sleep efficiency measured by an electronic device and by PSQI questionnaire, the average deviation between them is equal to 14% for all test persons and 9% in case of exclusion of subjects 7 and 8 for the reasons of missing records for more than the half of nights of the measurement period. 3.7. Comparison of objective and subjective measurement As seen in previous sections, the results of objective measurements obtained using electronic devices and outcomes of questionnaires based subjective measurements are very similar for some characteristics, but they differ significantly for others. The mean and median deviation between objective and subjective measurement, as well as standard deviation (SD) are presented in Table 3. Table 3. Difference between objective and subjective measurements. Characteristic Mean Meadian SD Time going to bed (hh:mm) 00:31 00:26 00:20 Total sleep time (TST) (hh:mm) 01:17 01:00 00:59 Total in bed time (TIB) (hh:mm) 00:51 00:43 00:40 Sleep efficiency (SE) 14% 9% 10% Lecture Notes in Electrical Engineering 627 Sergio Saponara Alessandro De Gloria Editors Applications in Electronics Pervading Industry, Environment and Society APPLEPIES 2019 [email protected] Editors Sergio Saponara DII University of Pisa Pisa, Italy Alessandro De Gloria DITEN University of Genoa Genoa, Italy ISSN 1876-1100 ISSN 1876-1119 (electronic) Lecture Notes in Electrical Engineering ISBN 978-3-030-37276-7 ISBN 978-3-030-37277-4 (eBook) https://doi.org/10.1007/978-3-030-37277-4 ©Springer Nature Switzerland AG 2020 This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This Springer imprint is published by the registered company Springer Nature Switzerland AG The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland Chapter 38 Recognizing Breathing Rate and Movement While Sleeping in Home Environment Maksym Gaiduk, Ralf Seepold, Natividad Martínez Madrid, Simone Orcioni and Massimo Conti Abstract The recovery of our body and brain from fatigue directly depends on the quality of sleep, which can be determined from the results of a sleep study. The classification of sleep stages is the first step of this study and includes the measurement of vital data and their further processing. The non-invasive sleep analysis system is based on a hardware sensor network of 24 pressure sensors providing sleep phase detection. The pressure sensors are connected to an energy-efficient microcontroller via a system-wide bus. A significant difference between this system and other approaches is the innovative way in which the sensors are placed under the mattress. This feature facilitates the continuous use of the system without any noticeable influence on the sleeping person. The system was tested by conducting experiments that recorded the sleep of various healthy young people. Results indicate the potential to capture respiratory rate and body movement. M. Gaiduk (B)·R. Seepold HTWG Konstanz, Alfred-Wachtel-Str. 8, 78462 Konstanz, Germany e-mail: [email protected] R. Seepold e-mail: [email protected] M. Gaiduk University of Seville, Av. Reina Mercedes s/n, 41012 Seville, Spain R. Seepold ·N. Martínez Madrid Department of Information and Internet Technology, Sechenov University, Bolshaya Pirogovskaya st., 119435 Moscow, Russian Federation e-mail: [email protected] N. Martínez Madrid Reutlingen University, Alteburgstr. 150, 72762 Reutlingen, Germany S. Orcioni ·M. Conti Department of Information Engineering, Università Politecnica delle Marche, Via Brecce Bianche, 12, 60131 Ancona, Italy e-mail: [email protected] M. Conti e-mail: [email protected] © Springer Nature Switzerland AG 2020 S. Saponara and A. De Gloria (eds.), Applications in Electronics Pervading Industry, Environment and Society, Lecture Notes in Electrical Engineering 627, https://doi.org/10.1007/978-3-030-37277-4_38 333 Conference Chairs Mohamad Sawan, Westlake University Carolyn McGregor AM, Ontario Tech University Atam Dhawan, NJIT Program Chairs Benoit Gosselin, Université Laval Joaquin Azpiroz Leehan, Universidad Autonoma Metropolitana Olaf Doessel, Karlsruhe Institute of Technology KIT Dominique Durand, Case Western Reserve University Indexed in PubMed® and MEDLINE®, Products of the United States National Library of Medicine  Abstract— This document presents a new complete standalone system for a recognition of sleep apnea using signals from the pressure sensors placed under the mattress. The developed hardware part of the system is tuned to filter and to amplify the signal. Its software part performs more accurate signal filtering and identification of apnea events. The overall achieved accuracy of the recognition of apnea occurrence is 91%, with the average measured recognition delay of about 15 seconds, which confirms the suitability of the proposed method for future employment. The main aim of the presented approach is the support of the healthcare system with the cost-efficient tool for recognition of sleep apnea in the home environment. Clinical Relevance— This approach can be used for continuous non-obtrusive apnea monitoring in the home environment. I. INTRODUCTION Humans spend a big part of everyday sleeping. Sleep is unavoidable for the normal functioning of the human organism. To restore our body and brain for daily life, it has to be healthy. However, for healthy sleep, not only the duration of sleep is essential, but its quality [1]. The usual approach for the analysis of sleep behavior is a study in the sleep laboratory. In this case, the overnight polysomnography (PSG) according to the guidelines of the American Academy of Sleep Medicine (AASM) is applied [2], which is very accurate but time and cost consuming procedure. PSG is a golden standard for a sleep study, which includes attaching several sensors to the human body for the measurement, among other things of EEG, ECG, EMG, and EOG signals [3]. Another critical point is that traditional PSG study is typically executed in a sleep laboratory and not in a home environment. Therefore, the continuous observation of a patient at home is almost not possible to organize. For this reason, cost-efficient systems for use in the home * This research was partially funded by the EU Interreg V-Program "Alpenrhein-Bodensee-Hochrhein": Project "IBH Living Lab Active and Assisted Living", grants ABH40, ABH41 and ABH66 and by the Spanish Ministry of Science, Innovation and Universities: EDITH Research Project (PGC2018-102145-B-C21,C22 (AEI/FEDER, UE)). M. Gaiduk is with the Ubiquitous Computing Lab at HTWG Konstanz, Alfred-Wachtel-Str. 8, 78462 Konstanz, Germany (phone: +49 7531 206- 703; email: [email protected]) and University of Seville, Avda. Reina Mercedes s/n, Seville, Spain M. Conti and S. Orcioni are with the Department of Information Engineering, Università Politecnica delle Marche via brecce bianche 12, I- 60131 Ancona, Italy R. Seepold is with the Ubiquitous Computing Lab at HTWG Konstanz, Alfred-Wachtel-Str. 8, 78462 Konstanz, Germany and the Department of environment could support sleep experts in analyzing a sleep quality, even if the accuracy of measurements in a home environment cannot achieve the level of PSG in a sleep laboratory. In [4], a review of actual health monitoring systems using sensors on bed or cushion was done. Sleep apnea is one of the sleep disorders, which is present by 3% to nearly 50% of the population, depending on sex and age [5,6]. It has different effects on our everyday life; for example, the daytime sleepiness was reported by about 25% of patients with obstructive sleep apnea (OSA) [7]. Fatigue or unrestful sleep was indicated by a higher amount of persons [7]. II. STATE OF THE ART Recognition of OSA is a topic of several scientific publications. Different approaches are presented in the following shortlisting of actual research on this topic. The sound was analyzed in [8] to recognize the OSA/events. It was processed using Voice Activity Detection (VAD) algorithm, which has measured the energy of the acoustic signal during the breathing and holding the breath. The accuracy of about 97% for the silence phases lasting 15s or longer in the quiet environment using the professional microphone was achieved in the experiment with 50 participating persons. In [9], the signal from bio-radar placed about 2,5 m from the test persons was used as the input of the system. The wavelet information entropy spectrum is a base for the algorithm implementation. The achieved accuracy for ten participating test subjects was equal to 93,1% comparing to the Polysomnography measurements. In a further research of the same group, they have applied the developed algorithm on the PSG respiratory recordings instead of bio/radar and achieved an accuracy of 96,1% for the detection of apnea events [10]. Information and Internet Technology at I.M. Sechenov First Moscow State Medical University, Moscow, Russia (email: [email protected]) T. Penzel is with the Interdisciplinary Sleep Medicine Center, Charité Universitätsmedizin Berlin, Charitéplatz 1, 10117 Berlin, Germany and International Clinical Research Center, St. Anne's University Hospital Brno, Brno, Czech Republic (email: [email protected]) N. Martínez Madrid is with the IoT Lab at Reutlingen University, Alteburgstr. 150, 72762 Reutlingen, Germany and the Department of Information and Internet Technology at I.M. Sechenov First Moscow State Medical University, Moscow, Russia (email: [email protected]) J.A. Ortega is with the University of Seville, Avda. Reina Mercedes s/n, Seville, Spain (email: [email protected]) Embedded system for non-obtrusive sleep apnea detection* Maksym Gaiduk, Student Member, IEEE, Simone Orcioni, Senior Member, IEEE, Massimo Conti, Member, IEEE, Ralf Seepold, Senior Member, IEEE, Thomas Penzel, Senior Member, IEEE, Natividad Martínez Madrid, Member, IEEE, Juan A. Ortega © 2020 IEEE. Reprinted, with permission, from Maksym Gaiduk et al., Embedded system for non-obtrusive sleep apnea detection, 2020 42nd Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC), July 2020 In reference to IEEE copyrighted material which is used with permission in this thesis, the IEEE does not endorse any of University of Seville's products or services. Internal or personal use of this material is permitted. If interested in reprinting/republishing IEEE copyrighted material for advertising or promotional purposes or for creating new collective works for resale or redistribution, please go to http://www.ieee.org/publications_standards/publications/rights/rights_link.html to learn how to obtain a License from RightsLink. 133 PART IV FINAL REMARKS APPENDIX A ___________________________________________________________________________ 140 A.1.2 Other scientific articles (incl. conference proceedings) Maksym Gaiduk has presented his research findings in international conferences over 20 times. In order to maintain a clear overview, the conferences are not listed additionally, as the contributions were also published in the corresponding proceedings and are listed below in the list of publications. 1. M. Gaiduk, R. Seepold, and N. Martínez Madrid, “Classification of sleep stages: commonly used methods and main aims for the improvement” In: Analysis of Biometric Parameters to detect relationship between stress and sleep quality (AnBiPa 2016), November 4, 2016, Università Politecnica delle Marche, Ancona, Italy, 2016, pp. 15–17. 2. A. Klein, N. Martínez Madrid, R. Seepold, and M. Gaiduk, “Sleep phase identification based on non-invasive recordings,” in Analysis of Biometric Parameters to detect relationship between stress and sleep quality (AnBiPa 2016), November 4, 2016, Università Politecnica delle Marche, Ancona, Italy, 2016, pp. 11–13. 3. M. Gaiduk, I. Kuhn, R. Seepold, J. A. Ortega, and N. Martínez Madrid, “A sensor grid for pressure and movement detection supporting sleep phase analysis,” in Bioinformatics and Biomedical Engineering, 5th International Work-Conference (IWBBIO 2017), Granada, Spain, April 26–28, 2017, Proceedings, Part II, (Lecture Notes in Computer Science; 10209), Cham: Springer, 2017, pp. 596–607. 4. M. Conti, S. Orcioni, N. Martínez Madrid, M. Gaiduk, and R. Seepold, “A review of health monitoring systems using sensors on bed or cushion,” in Bioinformatics and Biomedical Engineering, 6th International Work-Conference (IWBBIO 2018), 25th- 27th April 2018, Granada, Spain - (Lecture Notes in Computer Science; Vol. 10814), Cham: Springer, 2018, pp. 347–358. 5. M. Gaiduk, R. Seepold, S. Orcioni, M. Conti, and N. Martínez Madrid, “Non-invasive sleep analysis with intelligent sensors,” in Smart-Future-Living-Bodensee: Proceedings of International Workshop, 24. November 2017, Hochschule Konstanz, 2018, pp. 23–26. 6. M. Gaiduk, R. Seepold, and T. Penzel, „Sensor-Netz-basiertes System für die langfristige Schlafanalyse im häuslichen Umfeld“, Abstracts der 26. Jahrestagung der Deutschen Gesellschaft für Schlafforschung und Schlafmedizin e.V., „Schlaf ist Medizin!“, 11.–13. Oktober 2018, Nürnberg - (Somnologie, 2018, Vol. 22, Suppl. 1). 7. M. Gaiduk, R. Seepold, and E. Rodríguez De Trujillo, “Sleep position recognition in home environment,” in Fachtagung "Smart Day 2018", Organiser: Smart Home & Living Baden-Württemberg e.V., 7. Nov. 2018, Villingen-Schwenningen, 2018, pp. 21– 24. [Online]. Available: http://nbn-resolving.de/urn/resolver.pl?https://www.shlbw.de/fileadmin/documents/SmartDay/Tagungsband_Smart_Day_2018.pdf 8. M. Gaiduk, B. Vunderl, R. Seepold, J. A. Ortega, and T. Penzel, “Sensor-Mesh-Based System with Application on Sleep Study,” in Bioinformatics and Biomedical Engineering, 6th International Work-Conference (IWBBIO 2018), 25th-27th April 2018, Granada, Spain - (Lecture Notes in Computer Science; Vol. 10814), Cham: Springer, 2018, pp. 371–382. 9. S. Orcioni, M. Conti, N. Martínez Madrid, M. Gaiduk, and R. Seepold, “A review of health monitoring systems using sensors on beds or cushion,” in Smart-Future-Living- Bodensee: Proceedings of International Workshop, 24. November 2017, Hochschule Konstanz, 2018, pp. 45–48. ___________________________________________________________________________ 141 10. E. Rodríguez De Trujillo, M. Gaiduk, and R. Seepold, “An approach for a sleep tracking system,” in Informatics Inside connect(IT), Informatik-Konferenz an der Hochschule Reutlingen, 9. Mai 2018, Reutlingen: Hochschule Reutlingen, 2018, pp. 111–112. [Online]. Available: https://infoinside.reutlingen-university.de/wp-con- tent/uploads/2018/09/2018-Tagungsband-bw.pdf 11. E. Rodríguez De Trujillo, R. Seepold, and M. Gaiduk, “Position recognition algorithm using a two-stage pattern classification set applied in sleep tracking,” 1877-0509, vol. 126, pp. 1819–1827, 2018, doi: 10.1016/j.procs.2018.08.095. 12. R. Seepold, M. Gaiduk, T. Penzel, and N. Martínez Madrid, “Contactless detection of sleep phases with the help of regression analysis,” in: 40th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC'18), 18-21 July 2018, Hawaii Convention Center Honolulu, HI, USA, 2018. 13. M. Gaiduk, R. Seepold, N. Martínez Madrid, J. A. Ortega, and T. Penzel, “Non-inva- sives System für die kontinuierliche Schlafanalyse,” 6. Ambient Medicine® Forum „Assistive Technik für selbstbestimmtes Wohnen“, 19.-20. February 2019, Hochschule Kempten, Tagungsband, pp. 23–28, 2019. 14. M. Gaiduk, R. Seepold, and T. Penzel, “Softwaresystem zur Erkennung von Schlaf- und Wachzuständen basierend auf der Analyse von Herzfrequenz-, Atem- und Bewegungssignalen,” in Abstract der 27. Jahrestagung der Deutschen Gesellschaft für Schlafforschung und Schlafmedizin e. V., „Die innere Uhr“, 07.–09. November 2019, Hamburg (Somnologie, 2019, Vol. 23, Suppl. 1), 2019, S2-S2. 15. M. Gaiduk, R. Seepold, T. Penzel, J. A. Ortega, M. Glos, and N. Martínez Madrid, “Recognition of sleep/wake states analyzing heart rate, breathing and movement signals,” in 41st Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC 2019), 23-27 July 2019, Messe Berlin, Germany: IEEE, 2019, pp. 5712–5715. 16. M. Jurić, M. Gaiduk, and R. Seepold, “Influence of Illuminance on Sleep Onset Latency in IoT Based Lighting System Environment,” in Bioinformatics and Biomedical Engineering, 7th International Work-Conference (IWBBIO 2019), May 8-10, 2019, Granada, Spain, Proceedings, Part I (Lecture Notes in Bioinformatics; 11465), Cham: Springer International Publishing, 2019, pp. 429–438. 17. S. Orcioni, R. Pellegrini, R. Seepold, M. Gaiduk, N. Martínez Madrid, and M. Conti, “mHealth for therapeutic adherence,” in EMBC Workshop "Telemedicine and Telemonitoring in AAL Home Environments", 41st Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC 2019), 23 July 2019, Messe Berlin, Germany, M. Conti, G. Lebedev, N. Martínez Madrid, S. Orcioni, and R. Seepold, Eds., Reutlingen: Hochschule Reutlingen, 2019, pp. 25–28. 18. E. Rodríguez De Trujillo, R. Seepold, M. Gaiduk, N. Martínez Madrid, S. Orcioni, and M. Conti, “Embedded system to recognize movement and breathing in assisted living environments,” in Applications in Electronics Pervading Industry, Environment and Society: APPLEPIES 2018, 26. - 27. September 2018, Pisa, Italy, (Lecture Notes in Electrical Engineering; Vol. 550), Cham: Springer International Publishing, 2019, pp. 391–397. 19. M. Sandybekov, C. Grabow, M. Gaiduk, and R. Seepold, “Posture tracking using a machine learning algorithm for a home AAL environment,” Intelligent Decision Technologies 2019, Proceedings of the 11th KES International Conference on Intelligent Decision Technologies (KES-IDT 2019) - Volume 2 ; (Smart Innovation, Systems and Technologies ; 143), pp. 337–347, 2019, doi: 10.1007/978-981-13-8303-8_31. APPENDIX A ___________________________________________________________________________ 142 20. R. Seepold, M. Conti, N. Martínez Madrid, C. Aironi, S. Orcioni, and M. Gaiduk, "Die Ballistokardiographie misst die Herzfrequenz anhand der mechanischen Körperschwingungen aus der Herzbewegung.," Somnologie, 2020/10/01 2020, doi: 10.1007/s11818-020-00267-4. 21. R. Seepold, M. Gaiduk, and N. Martínez Madrid, “Personalized health service in assistive environments and telemonitoring of sleep patterns,” in EMBC Workshop "Telemedicine and Telemonitoring in AAL Home Environments", 41st Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC 2019), 23 July 2019, Messe Berlin, Germany, M. Conti, G. Lebedev, N. Martínez Madrid, S. Orcioni, and R. Seepold, Eds., Reutlingen: Hochschule Reutlingen, 2019, pp. 15–17. 22. R. Seepold, M. Gaiduk, J. A. Ortega, M. Conti, S. Orcioni, and N. Martínez Madrid, “Home hospital e-health centers for barrier-free and cross-border telemedicine,” Intelligent Decision Technologies 2019, Proceedings of the 11th KES International Conference on Intelligent Decision Technologies (KES-IDT 2019) - Volume 1; (Smart Innovation, Systems and Technologies; 143), pp. 307–316, 2019, doi: 10.1007/978-981- 13-8303-8_28. 23. R. Seepold, K. Stuburić, M. Gaiduk, N. Martínez Madrid, and T. Penzel, "Aufwandsbegrenztes maschinelles Lernen zur Bestimmung von Schlafstadien," Somnologie, 2020/10/01 2020. 24. M. Conti, C. Aironi, S. Orcioni, R. Seepold, M. Gaiduk, and N. Martinez Madrid, “Heart rate detection with accelerometric sensors under the mattress,” in 42nd Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC 2020), 20-24 July 2020, Montreal, QC, Canada, virtual, 2020, pp. 4063–4066. 25. M. Gaiduk et al., “Embedded system for non-obtrusive sleep apnea detection,” in 42nd Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC 2020), 20-24 July 2020, Montreal, QC, Canada, virtual, 2020, pp. 2776– 2779. 26. M. Gaiduk, R. Seepold, N. Martínez Madrid, S. Orcioni, and M. Conti, “Recognizing breathing rate and movement while sleeping in home environment,” in Applications in Electronics Pervading Industry, Environment and Society: APPLEPIES 2019, 11. - 13. September 2019, Pisa, Italy, (Lecture Notes in Electrical Engineering; Vol. 627), Cham: Springer, 2020, pp. 333–339. 27. H.-P. Hutter, A. Darvishy, S. Roth, S. Gäumann, H. Kaspar, T. Thimm, M. Gaiduk, S. Evans, and M. Rosenberg “Service Design for Accessible Tourism,” in Universal Access in Human-Computer Interaction: Applications and Practice, HCII 2020, 22nd International Conference on Human-Computer Interaction, 19-24 July 2020, Copenhagen, Denmark, virtual, (Lecture Notes in Computer Science, vol. 12189), Cham: Springer, 2020, pp. 409–419. 28. M. Gaiduk, R. Seepold, and T. Penzel, "Kontaktloses System zur Erkennung von Apnoe," Somnologie, 2020/10/01 2020, doi: 10.1007/s11818-020-00267-4 29. L. Weber, M. Gaiduk, W. D. Scherz, and R. Seepold, “Cardiac Abnormality Detection in 12-lead ECGs With Deep Convolutional Neural Networks Using Data Augmentation,” in Computing in Cardiology (CinC 2020, September 13th-16th, Rimini, Italy, virtual): IEEE, 2020, p. 4. 30. M. Gaiduk, D. Wehrle, R. Seepold, and J. A. Ortega, "Non-obtrusive system for overnight respiration and heartbeat tracking," Procedia Computer Science, vol. 176, pp. 2746-2755, 2020/01/01/ 2020, doi: https://doi.org/10.1016/j.procs.2020.09.282 ___________________________________________________________________________ 143 31. Y. Shkilniuk, Á. S. Alarcón, M. Gaiduk, R. Seepold, and N. Martínez Madrid, "Analysis of Survey Tools for Recommender Systems in the Selection of Ambient Assisted Living Technologies," Procedia Computer Science, vol. 176, pp. 2332-2340, 2020/01/01/ 2020, doi: https://doi.org/10.1016/j.procs.2020.09.294 32. M. Gaiduk, R. Seepold, J. A. Ortega, and N. Martínez Madrid, "Comparison of sleep characteristics measurements: a case study with a population aged 65 and above," Procedia Computer Science, vol. 176, pp. 2341-2349, 2020/01/01/ 2020, doi: https://doi.org/10.1016/j.procs.2020.09.297 33. K. Stuburić, M. Gaiduk, and R. Seepold, "A deep learning approach to detect sleep stages," Procedia Computer Science, vol. 176, pp. 2764-2772, 2020/01/01/ 2020, doi: https://doi.org/10.1016/j.procs.2020.09.280. 34. Y. Shkilniuk, M. Gaiduk, and R. Seepold, “Unobtrusive Accelerometer-Based Heart Rate Detection,” in Applications in Electronics Pervading Industry, Environment and Society: APPLEPIES 2020, 19. - 20. November 2020, Lecture Notes in Electrical Engineering; Vol. 738, Cham: Springer, 2021, pp. 49–54. 35. M. Gaiduk et al., “A Comparison of Objective and Subjective Sleep Quality Measurement in a Group of Elderly Persons in a Home Environment,” in Applications in Electronics Pervading Industry, Environment and Society: APPLEPIES 2020, 19. - 20. November 2020, virtual, Lecture Notes in Electrical Engineering; Vol. 738, Cham: Springer, 2021, pp. 286–291. 36. S. Orcioni, F. Di Nardo, S. Fioretti, M. Conti, R. Seepold, M. Gaiduk, and N. Martínez Madrid, "Preliminary results of homomorphic deconvolution application to surface EMG signals during walking," Procedia Computer Science, vol. 192, pp. 3272-3280, 2021/01/01/ 2021, doi: https://doi.org/10.1016/j.procs.2021.09.100 37. L. Weber, M. Gaiduk, R. Seepold, N. Martínez Madrid, M. Glos, and T. Penzel, “Enhancing Current Cardiorespiratory-based Approaches of Sleep Stage Classification by Temporal Feature Stacking,” 43rd Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC 2021), Oct 31 - Nov 04 2021, virtual, pp. 5518–5522, 2021, doi: 10.1109/EMBC46164.2021.9630743. 38. S. Orcioni, R. Pellegrini, R. Seepold, M. Gaiduk, N. Martínez Madrid, and M. Conti, “Medication adherence supported by mHealth and NFC,” 2352-9148, vol. 23, pp. 1– 12, 2021, doi: 10.1016/j.imu.2021.100552. 39. M. Gaiduk, R. Seepold, and L. Weber, “Digital technologies for accessible tourism,” in Grenzüberschreitende Reallabore für Assistenztechnik: Beiträge zum uDay XIX, 24.6.2021, Österreich, Veranstalter: IBH Living Lab AAL Active & Assisted Living, G. Kempter and W. Ritter, Eds., Germany: Pabst Science Publishers, 2021, pp. 119–127. 40. R. Seepold, M. Gaiduk, and W. D. Scherz, “Home Health Systems: conception and implementation results,” in Grenzüberschreitende Reallabore für Assistenztechnik: Beiträge zum uDay XIX, 24.6.2021, Österreich, Veranstalter: IBH Living Lab AAL Active & Assisted Living, G. Kempter and W. Ritter, Eds., Germany: Pabst Science Publishers, 2021, pp. 128–137. 41. L. Weber, M. Gaiduk, R. Seepold, and T. Penzel, “Einsatz rekurrenter neuronaler Netze als unabhängige Verbesserung bereits trainierter Algorithmen zur Erkennung von Schlafphasen,” in Abstracts der 29. Jahrestagung der Deutschen Gesellschaft für Schlafforschung und Schlafmedizin e. V., "Schlaf und Psyche", 28.-30. Oktober 2021, virtuell (Somnologie, 2021, Vol. 25, Suppl. 1), 2021, S29-S29. 42. M. Gaiduk, R. Seepold, and N. Martínez Madrid, “Development of Health Systems for Home Environments with the Focus on Sleep Monitoring,” in Grenzüberschreitende APPENDIX A ___________________________________________________________________________ 144 Reallabore für Assistenztechnik: Beiträge zum uDay XIX, 24.6.2021, Österreich, Veranstalter: IBH Living Lab AAL Active & Assisted Living, G. Kempter and W. Ritter, Eds., Germany: Pabst Science Publishers, 2021, pp. 138–144. 43. M. Gaiduk, R. Seepold, N. Martínez Madrid, and T. Penzel, “Identifikation von Schlaf- und Wachzuständen durch die Auswertung von Atem- und Bewegungssignalen,” in Abstracts der 29. Jahrestagung der Deutschen Gesellschaft für Schlafforschung und Schlafmedizin e. V., "Schlaf und Psyche", 28.-30. Oktober 2021, virtuell (Somnologie, 2021, Vol. 25, Suppl. 1), 2021, S30-S31. 44. M. Gaiduk, L. Weber, Á. Serrano Alarcón, R. Seepold, N. Martínez Madrid, S. Orcioni, and M. Conti, "Design of a sleep apnoea detection system for a home environment," in Procedia Computer Science, vol. 192, pp. 3225-3234, 2021, doi: https://doi.org/10.1016/j.procs.2021.09.095. A.2 Awards 1) In November 2021, Maksym Gaiduk was awarded as the author of the best article among young scientists at the international workshop "Social Innovation in Long-Term Care through Digitalization (LTC-2021)" in Ancona, Italy. 2) In 2021, the paper "Estimation of Sleep Stages Analyzing Respiratory and Movement Signals", published in Journal of Biomedical and Health Informatics, for which Maksym Gaiduk is the first and corresponding author, was selected as the best paper of July 2021 and third-best paper of the year at Escuela Técnica Superior de Ingeniería Informática (ETSII) of the University of Seville. A.3 Research exchanges During my PhD study period, I have spent several months on international research exchanges: 1) Sevilla (Spain). April 2017 (9 days). During this stay, several meetings with research groups of the School of Computer Engineering (ETSII) of Seville University took place to exchange the experience and develop common research directions. 2) Ancona (Italy). October 2017 (3 weeks). The aim of this stay was knowledge and experience exchange with research groups from Università Politecnica delle Marche. 3) Bratislava (Slovakia). October – November 2017 (1 week). This stay aimed to exchange knowledge and experience with research groups from the University of Bratislava. ___________________________________________________________________________ 145 4) Bucharest (Romania). September 2018 (3 days). The three-day meeting with the representatives of the research groups of universities from 5 countries. 5) Sofia (Bulgaria). August 2019 (4 days). The four-day meeting with the representatives of the research groups of universities from 5 countries. 6) Ancona (Italy). September – December 2019 (3 months). I was hosted by Prof. Massimo Conti in Università Politecnica delle Marche. My stay was supported by the University of Seville. During this exchange, new developments were performed, and a research study with 20 participants using the developed by us system was designed and conducted. My usual place of residence was in Germany, and my regular workplace was at the HTWG Konstanz (Germany). However, during this time, I had had numerous research stays in Seville (Spain) with a total duration of over two months. A.4 Projects participation 1. Title: "IBH AAL: Home Health Living Lab (HHLL)" Main Researcher: Ralf Seepold Granting Entity: EU Interreg V-Program "Alpenrhein-Bodensee-Hochrhein" Period: 2018-2021 Reference: ABH066 2. Title: "IBH Living Lab Active and Assisted Living" Main Researcher: Guido Kempter Granting Entity: EU Interreg V-Program "Alpenrhein-Bodensee-Hochrhein" Period: 2016-2021 Reference: ABH040, ABH041 3. Title: “Development of Online Learning Environment for e-Health (DOOLEE)” Granting Entity: The European Commission: ERASMUS + Program Period: 2018-2020 APPENDIX A ___________________________________________________________________________ 146 Reference: 2017-1-BG01-KA203-036310 4. Title: Evolucionando hacia los Gemelos Digitales en el Ámbito de la Salud (EDITH) Main Researcher: Juan Antonio Ortega Ramírez Granting Entity: Spanish Ministry of Science, Innovation and Universities. (Ministerio de Ciencia, Innovación y Universidades: Plan Estatal de Investigación Científica y Técnica y de Innovación 2017-2020 Generación Conocimiento - Proyectos I+D+I) Period: 2020-2021 Reference: PGC2018-102145-B-C21 5. Title: ZIM project “Sleep Lab at Home (SLaH)” Granting Entity: German Federal Ministry for Economic Affairs and Energy Period: 2020-2023 Reference: ZF4825301AW9 6. Title: “Non-invasive system for measuring parameters relevant to sleep quality” Main Researcher: Ralf Seepold Granting Entity: Carl Zeiss Foundation Period: 2021-2024 Reference: P2019-03-003 A.5 Open Science One of the important tasks of science is to use knowledge to improve the condition of human beings. In this sense, the accessibility of the scientific results achieved is of great importance. In accordance with this, an effort has been made to make the research results accessible to everyone as far as possible, resulting ___________________________________________________________________________ 147 in a large part of the publications being published according to the open-access principle, including one of the JCR journal articles. A.6 Others 1) Maksym Gaiduk was actively involved in developing the cooperation between the Ubiquitous Computing Lab at HTWG Konstanz and research group IDINFOR at the University of Seville. This cooperation has led, among other things, to several research staff exchanges, numerous organised Special Sessions at international conferences, and joint investigations. 2) Maksym Gaiduk was one of the organising committee members of the International Workshop "Smart-Future-Living-Bodensee" in Konstanz (Germany) in November 2017. 3) Maksym Gaiduk was a member of the organisation committee of GRSS-2018 in September 2018. 4) Maksym Gaiduk was a chairman of Special Session "Telemedicine for Smart Homes and Remote Monitoring" at the international conference IWBBIO 2019. 5) Maksym Gaiduk has been the deputy project leader of the international project "IBH AAL: Home Health Living Lab (HHLL)" over a period of 7 months in the year 2019. 6) Maksym Gaiduk is a reviewer for the JCR-indexed scientific journal "Transactions on Biomedical Engineering" (Q1). 7) Maksym Gaiduk is a reviewer for the JCR-indexed scientific journal "IEEE Access" (Q1). 8) Supervising of 4 Bachelor and Master Theses was done by Maksym Gaiduk. 148 149 BIBLIOGRAPHY BIBLIOGRAPHY ___________________________________________________________________________ 156 [57] L. Fiorillo et al., “Automated sleep scoring: A review of the latest approaches,” Sleep medicine reviews, vol. 48, p. 101204, 2019, doi: 10.1016/j.smrv.2019.07.007. [58] O. Faust, H. Razaghi, R. Barika, E. J. Ciaccio, and U. R. Acharya, “A review of automated sleep stage scoring based on physiological signals for the new millennia,” Computer methods and programs in biomedicine, vol. 176, pp. 81–91, 2019, doi: 10.1016/j.cmpb.2019.04.032. [59] G. D. Garson, Logistic Regression: Binary & Multinomial: 2016 Edition. Asheboro, NC: Statistical Associates Publishers, 2016.