Assessment of retinal vein pulsation through video-ophthalmoscopy and simultaneous biosignals acquisition

Research Article Vol. 14, No. 6 / 1 Jun 2023 / Biomedical Optics Express 2645 Assessment of retinal vein pulsation through video-ophthalmoscopy and simultaneous biosignals acquisition RADIM KOLAR,1TOMAS VICAR,1JIRI CHMELIK,1ROMAN JAKUBICEK,1,* JAN ODSTRCILIK,1EVA VALTEROVA,1MICHAL NOHEL,1KAROLINA SKORKOVSKA,2,3 AND RALF P. TORNOW4 1Department of Biomedical Engineering, Faculty of Electrical Engineering and Communication, Brno University of Technology, Brno, Czech Republic 2Deparment of Ophthalmology and Optometry, St. Ann University Hospital, Brno, Czech Republic 3Department of Ophthalmology and Optometry, Masaryk University, Brno, Czech Republic 4 Department of Ophthalmology, Friedrich-Alexander-University Erlangen–Nürnberg, Erlangen, Germany *[email protected] Abstract: The phenomenon of retinal vein pulsation is still not a deeply understood topic in retinal hemodynamics. In this paper, we present a novel hardware solution for recording retinal video sequences and physiological signals using synchronized acquisition, we apply the photoplethysmographic principle for the semi-automatic processing of retinal video sequences and we analyse the timing of the vein collapse within the cardiac cycle using of an electrocardiographic signal (ECG). We measured the left eyes of healthy subjects and determined the phases of vein collapse within the cardiac cycle using a principle of photoplethysmography and a semi-automatic image processing approach. We found that the time to vein collapse (T vc ) is between 60 ms and 220 ms after the R-wave of the ECG signal, which corresponds to 6% to 28% of the cardiac cycle. We found no correlation between T vc and the duration of the cardiac cycle and only a weak correlation between T vc and age (0.37, p = 0.20), and T vc and systolic blood pressure (-0.33, p = 0.25). The Tvc values are comparable to those of previously published papers and can contribute to the studies that analyze vein pulsations. Published by Optica Publishing Group under the terms of the Creative Commons Attribution 4.0 License. Further distribution of this work must maintain attribution to the author(s) and the published article’s title, journal citation, and DOI. 1. Introduction The retina is the only human tissue that allows direct non-invasive imaging of the microvascular circulation. Therefore, pulsatile changes induced by the cardiac cycle can be observed throughout the retina, mainly at the optic nerve head (ONH) and in the peripapillary region. These pulsatile phenomena include venous pulsation (often referred to as spontaneous venous pulsation, SVP), arterial pulsation (including changes in artery diameter and bending), and changes in light absorption of a specific wavelength due to changes at the microcapillary and vascular level (i.e. retinal photoplethysmography, PPG) [1,2]. The pulsating nature of these phenomena is primarily driven by the cardiac cycle. However, other factors also contribute and influence retinal pulsations. Since the intraocular space is close to the brain and the dura of the brain extends as a sheath around the ONH, cerebrospinal fluid pressure (CSFP) influences some of these retinal phenomena. Another variable is intraocular pressure (IOP), which can primarily influence SVP and PPG pulsations. Both pressures (CSFP and IOP) are not constant but also pulsate (see, e.g., IOP pulsations [3] or CSFP pulsations [4]). In this paper, we focus on the measurement of vein pulsation and its analysis. These pulsations are often referred to as spontaneous vein pulsations, although they are not actually “ spontaneous ” #486052 https://doi.org/10.1364/BOE.486052 Journal © 2023 Received 19 Jan 2023; revised 19 Apr 2023; accepted 19 Apr 2023; published 12 May 2023 Research Article Vol. 14, No. 6 / 1 Jun 2023 / Biomedical Optics Express 2646 because they are controlled by temporal changes in the translaminar pressure gradient [5]. Therefore, the heart is the “ pulse generator ” , and the magnitude and phase of IOP and CSFP create the condition for the vein to collapse. Another factor that can influence SVP is venous capacitance [5], which defines the stretching ability of the vein wall. The pressure gradient is most evident at the ONH, and therefore SVPs are observed primarily in this region. Since the translaminar pressure gradient is the result of the difference between IOP and CSFP, which are influenced by the cardiac cycle, it can be assumed that SVPs are also dependent on the cardiac cycle. This has been confirmed by several studies [6–14]. However, there is one question that remains open: what is the timing or phase of SVP during the cardiac cycle? Some studies have been published on this topic [7–9,11,15], but the results are not uniform. The importance of this parameter could arise when considering ocular diseases associated with changes in blood flow (e.g. retinal vein occlusion, glaucoma) or when considering cardiovascular diseases such as carotid stenosis, as the latter affects retinal blood flow [16]. Furthermore, SVP can be used to estimate intracranial pressure [17], where the SVP phase may also play a significant role. Finally, an accurate measurement of the SVP phase could help to better understand its origin and possible future applications. In SVP analysis, an appropriate spatial and temporal resolution must be used. Spatial resolution is usually not a problem because some commercially available retinal imaging solutions are used. However, the temporal resolution should be high enough to capture potential variability within the subject and between subjects in the SVP phase. Several studies published on this topic have revealed important information about SVP [5,15,18–21], but many of them use a relatively small temporal sampling rate (typically around 10 fps). Some studies have used a retinal vascular analyzer (RVA, Imedos UG, Jena, Germany), which allows the acquisition of retinal image sequences with a sufficient frame rate (around 25 fps), but the light intensity is relatively high, which is not comfortable for subjects. Here, we present a novel hardware solution for acquiring retinal video sequences and physiological signals using trigger pulses that preserve the possibility of synchronous analysis. The advantage of this solution is that it is relatively easy to setup and use in a clinical setting. In addition, it uses low-intensity pupil illumination to make the measurement more comfortable for the subject. Last but not least, it works with both dilated and non-dilated pupils. Furthermore, we utilize a photoplethysmographic principle to analyze the pulsation, which does not need an exact extraction of the vein diameter. We tested our setup in a group of healthy subjects and analyzed the SVP phase during the cardiac cycle. Finally, the results are discussed and compared with other published work on SVP analysis. 2. Materials and methods 2.1. Acquisition system The acquisition system consists of three main parts (Fig. 1): 1. Video-ophthalmoscope (VO) – The monocular video-ophthalmoscope is based on our previous research on binocular video-ophthalmoscopy [1,2]. The setup of the version used in the previous work was modified to a monocular version with the possibility of multispectral measurement [22]. This device acquires retinal video of the optic nerve head and the peripapillary area with a field of view of 20 °× 17 ° (1224 × 970 pixels). This corresponds to an approximate area of 6 × 5 mm in the retina for a subject with an axial length of 24 mm. The frame rate of the CMOS camera (UI-3060 Rev 2, USB 3.0, IDS Imaging Development Systems GmbH, Germany) was set at 25 fps (camera exposure time approximately 40 ms) and the light power generated by an external light source (CoolLED pe-4000, CoolLED Ltd., UK) in the plane of the eye pupil was less than 15 µ W. This results in a retinal illumination of 50 µ W/cm 2 , which corresponds to 16.6% of the maximum Research Article Vol. 14, No. 6 / 1 Jun 2023 / Biomedical Optics Express 2647 exposure limit according to [23]. The lengths of the video sequence are between 10 and 15 seconds. 2. Biosignal acquisition unit - The BiosignalPlux 4 channel unit (PLUX Wireless Biosignals S.A., Lisbon, Portugal) was used in this setup. We used an electrocardiographic (ECG) sensor, an ear photoplethysmography (PPG) sensor, and a respiratory band to record the corresponding signals. In this study, we used only the ECG signal. The fourth channel was used to record the trigger pulses (see the next paragraph). All signals were sampled at a sampling rate of 1000 Hz. 3. Synchronization unit - This unit consists of an Arduino platform which generates a square trigger signal to drive the camera acquisition. Each rising edge of this square signal triggers the frame acquisition. At the same time, this trigger signal is recorded by the BiosignalPlux unit together with the measured biosignals. This enables us to precisely determine the time of each frame within the biosignals. Thus, the retinal video and biosignals are precisely synchronized. VO optic Camera Sync unit BiosignalPlux Biosignals: ECG Ear PPG Respiratory Acqusition software Light source Fig. 1. Illustration of our proposed acquisition setup. 2.2. Subjects Thirteen healthy subjects were measured after pupil dilation (tropicamide drops, concentration of 0.25%). First, blood pressure (Omron, Intellisense) and intraocular pressure (Huwitz, non-contact tonometer HNT-1) were measured. The fundus image with a clinical fundus camera (Canon CR-1 with Canon EOS 40D). During measurement with a video-ophthalmoscope, we used various wavelengths [22], but here we analyzed video sequences, which were acquired with 525 nm. Light with this wavelength provides a good contrast of the blood vessels. This study followed the principles of the Declaration of Helsinki for research involving human subjects, and informed consent was obtained from all study participants. The measurement procedure was reviewed and approved by the Ethics Committee of Brno University of Technology (no. EK:03b/2021). Research Article Vol. 14, No. 6 / 1 Jun 2023 / Biomedical Optics Express 2648 2.3. Signal and image processing The ECG signal was processed to detect the positions of the R-wave in each cardiac cycle. We proposed an automatic detection that consists of filtering and robust local maxima detection. The signal was filtered with a median filter (size 10), high-frequency components were removed with a Gaussian filter ( σ= 12), and the baseline was removed by subtraction of a low-frequency signal obtained with a Gaussian filter ( σ= 200). Then, the robust local maxima detector with a peak threshold, minimal peak distance, and minimal peak prominence (minimal value drop between peaks) was used to detect the positions of R-waves. Here, the parameters were set adaptively (i.e. automatically) for each signal – threshold as (signal maximum) ⁄ 4, peak prominence as (signal maximum – signal minimum) ⁄ 2 and minimal distance as 0.545 seconds (maximum 110 beats per minute). This leads to the correct detection of all R-waves in all signals, which was confirmed by visual inspection as well. Similarly, the rising edges of the acquired synchronization square trigger signal were detected with the same local maxima detector, where the detector was applied to the first-order difference of this signal; however, detection without filtering and peak prominence was sufficient. A threshold half of the square wave magnitude was applied with a minimal peak distance of 0.6 × square wave period. The simultaneous acquisition of biosignals and video allows us to determine the temporal position of each R-wave in video recording. Thus, the frame that corresponds to the position of the R-wave can be determined. It should be noted that due to the different sampling rates for biosignals and video sequences, the exposure time of each frame (40ms) corresponds to 40 samples in the ECG signal. Retinal image sequences were processed in several steps corresponding to partial blocks in Fig. 2(see the flow diagram in Fig. 2): Fig. 2. The main flow diagram of the methodology; VP refers to vein pulsation. Frame-to-frame alignment – The previously published approach [22] was used to align each frame with respect to the first frame (i.e. reference). This approach corrects for translational and rotational movements using a combination of phase correlation and tracking of the center lines of the blood vessels. The registration error of this approach is 0.78 ± 0.67 pixels inside the optic disc and 1.39 ±0.63 pixels outside the optic disc, respectively [22]. Detection of distorted frames - Due to the eye movements, which may also occur during the frame exposure time, some frames can be blurred. Furthermore, blinking artifacts also cause significant distortion. These distorted frames cannot be restored and must be detected before analysis. Our previously developed method [24] for different data sets did not provide satisfactory results for these new sequences. Therefore, a new approach was used based on the similarity of the frames with their neighbors was used; specifically, the Euclidean distance (ED) between each frame and the median image from 71 surrounding frames was calculated and used for the detection of distorted frames. From EDs, distorted frames were detected as outliers using the generalized extreme Studentized deviate test [25], where most extreme values of ED are Research Article Vol. 14, No. 6 / 1 Jun 2023 / Biomedical Optics Express 2649 iteratively removed until outliers are not recognized with the test. This unsupervised automatic detection was refined manually by visual inspection of frames with higher ED values compared to their neighbors, which can contain weak distortion. Vein delineation - In the next step, only the pulsating part of the central retinal vein was manually delineated in ImageJ. A freehand tool was used to create one region on the vein during video playback. Vein pulsation signal extraction - The region of interest (ROI) from the previous step was then used to extract the average intensity from this ROI in each frame. This results in a signal representing the pulsation of the vein (referred to here as the VP signal, denoted I VP (n)). In this signal, each cardiac cycle can be clearly recognized (Fig. 3). The minimum in each cycle represents the lowest intensity in ROI, which corresponds to the maximum diameter of the vein (the vein creates a dark structure inside the ONH). In contrast, the maximum value for each cardiac cycle represents the collapse of the vein, because during this collapse, the diameter of the vein decreases (see Fig. 3). Fig. 3. a) An average image of a sequence with manually delineated ROI for vein pulsation analysis. b) Corresponding color fundus image. c) Vein pulsation signal extracted as an average value from ROI from each frame. The red curve represents the moving average signal, I AVG (n) d) Relative pulsatile signal, Irel VP(n) . e) Photoplethysmographic representation of vein pulsatile signal, I, VP(n). SVP signal interpolation – The indexes of the corrupted frames (Step 2) were used for interpolation of the corresponding values of VP signal from their neighbors (1D cubic interpolation). SVP signal processing –Reflected intensity from the retina can also be distorted byillumination intensity changes (caused mainly by small eye and head movements and the fluctuation of the pupil diameter), see Fig. 3(c). This affects especially the low-frequency part of the VP signal. To compensate for these changes, we computed the relative intensity Irel VP(n)as [2]: Irel VP(n)= IVP(n) IAVG(n), where IAVG(n) represents a low-frequency component estimated by a moving average filter with the length of the window equal to the 1.5 multiple of the average cardiac cycle length. The typical length of the impulse response was around N = 35 samples. This relative signal (see the example Research Article Vol. 14, No. 6 / 1 Jun 2023 / Biomedical Optics Express 2650 in Fig. 3(d)) fluctuates around 1 and allows us to easily define an alternative representation of vein pulsation as I, VP(n) [%]=100(1−Irel VP(n)) =100 (︃1−IVP(n) IAVG(n))︃. This representation can be interpreted as a photoplethysmographic signal, where the value is related to the volume of the corresponding part of the vein. Due to the minus sign in the equation, we now have a signal, where each minimum represents the minimum volume (i.e. vein collapse) and each maximum is the maximum vein volume during the cardiac cycle. An example of this signal is shown in Fig. 3(e). Selection of undistorted cardiac cycles – Even though we detected the distorted frames and used the interpolation, there can still be some parts of the VP signal, where the particular cardiac cycles are distorted. This is mainly due to the blinking of the eyes or large movements of the eye. These distorted cycles were manually excluded from the analysis. Finally, the number of undistorted cardiac cycles for each subject was typically between 5 and 11 cycles. Detection of vein collapse - Vein collapse is represented by a local minimum during each cardiac cycle in the VP signal. The position of each minimum was detected in a fixed temporal window (220 ms) after the R-wave. Visual inspection of each detected minima showed the correctness of each position. The length of the window was set according to an assumption based on the literature review (see Discussion). The plot of the simultaneously acquired ECG and VP signal is shown in Fig. 4, together with the detected R-waves and vein collapses. Fig. 4. a) Vein pulsation signal. b) ECG signal. The red vertical lines show the position of detected R-waves. The green vertical lines represent the time of the vein collapses. A small variation of the interval between the red and green lines can be observed. 3. Results 3.1. Correlation among clinical parameters and vein collapse time Scatter plots between T vc and other parameters are plotted (Fig. 5) and the correlation was evaluated using the Spearman correlation coefficient. We found a significantly higher correlation coefficient only for systolic BP (-0.33, p = 0.25) and age (0.37, p = 0.20). The remaining calculated Spearman correlation coefficients were below 0.2. Another not well-known fact is the relation between T vc and the duration of the cardiac cycle. The scatter plot in Fig. 6shows this relation, where each dot represents single cardiac cycle, Research Article Vol. 14, No. 6 / 1 Jun 2023 / Biomedical Optics Express 2651 Fig. 5. Scatter plots between vein collapse time (T vc ) on the x-axis and particular parameters on the y-axis: blood pressure (BP; systolic and diastolic), intraocular pressure (IOP), standard deviation of RR intervals (SDRR),heart rate and age. Fig. 6. The scatterplot of the TVC collapse time and the cardiac cycle length. Each color represents one of 13 subjects. Research Article Vol. 14, No. 6 / 1 Jun 2023 / Biomedical Optics Express 2652 and each colour represents one subject (in total 85 cardiac cycles from all 13 subjects were considered). The Spearman correlation coefficient was below 0.1, showing that these values are not correlated. 3.2. Vein collapse time analysis The T vc was determined for each undistorted cardiac cycle (according to the VP signal). These values were visualized as a boxplot for each subject and as a one violin plot for all subjects together (Fig. 7). Similarly, T vc was expressed in milliseconds and in percent with respect to the duration of the cardiac cycle. Fig. 7. a) Boxplot of T VC for each subject. b) Violin plot of T VC for all subjects. c) Boxplot of relative TVC for each subject. d) Violin plot of relative TVC for all subjects. Our setup and analysis also allow us to plot a single cardiac cycle represented by the VP signal and the ECG signal. This representation clearly shows the natural heart rate variability of the subject during the acquisition; see an example for one subject in Fig. 8. However, as shown above, no correlation was found between T vc and SDRR (representing this heart rate variability). Research Article Vol. 14, No. 6 / 1 Jun 2023 / Biomedical Optics Express 2653 Fig. 8. An example of several cardiac cycles represented by the ECG signal (a) and the VP signal (b) of one subject. The heart rate variability is clearly visible as different lengths of each heartbeat in the ECG signal. The vertical dashed line represents the time of the R-wave. The VP signals for each cardiac cycle were vertically aligned to zero at the time of the peak of the R wave. 4. Discussion Our study investigates vein pulsation and vein collapse phase in relation to the cardiac cycle, represented by the ECG signal in healthy subjects and using the photoplethysmographic principle. The results clearly show (Fig. 7) that vein collapse occurs in a systolic phase of the cardiac cycle in healthy subjects. To compare our findings and get a more complex view, including also the IOP cycle and the retinal artery cycle, we graphically summarize this in Fig. 9. We did not measure IOP pulsation; however, this IOP ocular cycle begins later than the cardiac cycle due to the time needed to induce changes in IOP after cardiac contraction and aortic valve. It was reported [25] that IOP cycle begins at 71 ° of the ECG cycle and peaks at 220 ° . The relationship between retinal arterial expansion and ECG can also be found in the same study [26], where the flow cycle in the central retinal artery begins at 46 ° and peaks at 111 ° of the cardiac cycle. The same comparison approach was also used by Levin et al. [5]. The relation between the cardiac cycle, IOP cycle and artery expansion cycle (also referred to as the ocular circulatory cycle) is shown in Fig. 9as blue, orange and green horizontal bars, respectively. This enables us to compare our results with more published results using IOP or artery expansion, not only with ECG signals. The boxes in Fig. 9represent the central retinal vein maximum diameter phase (CRV max ) or central retinal vein minimum diameter phase (CRV max ) published by various authors. These