Malaria Parasites Detection System using Machine Learning Techniques

101 Views - 7 Downloads


Malaria Parasites Detection System using Machine Learning Techniques #0 Malaria Parasites Detection System using Machine Learning Techniques #1 Malaria Parasites Detection System using Machine Learning Techniques #2 Malaria Parasites Detection System using Machine Learning Techniques #3 Malaria Parasites Detection System using Machine Learning Techniques #4


1 Malaria Parasites Detection System using Machine Learning Techniques S M Raju1*, Ali Mohammad Tarif2, Mishkat Nur Rahman3, Md. Shariful Islam4 Abstract—The purpose of this project is to build a detection system for malaria parasites in images. This would avoid some of the current obstacles of malaria diagnosis such as the lack of expert technicians in low-resource areas. In contrast to most of the previous work we use thick blood smears instead of thin smears. These are the preferred ones for diagnosis, although they make the task harder. We use Convolutional Neural Networks, a particular type of deep neural network, to classify individual patches as containing parasites or not. Then, we use computer vision techniques such as non-maxima suppression to integrate the results and locate each parasite. The results are compared to those obtained with Extremely Randomized Trees and hand-designed features as used in previous work on this dataset (Quinn et al, 2014 [1]). Our results are notably better than the models proposed previously. For a recall of 90% we have obtained a gain in the precision from 37% to 78%, which is enough to already improve the work conditions of laboratory technicians working on malaria diagnosis. Furthermore, the system can be easily extended to the detection of other parasites or blood components without much effort, so several tests may be run at the same time on the blood smears. Index Terms— Convolutional Neural Networks, Diagnosis, Malaria, Parasite. I. INTRODUCTION T HE gold standard test for malaria is the hundred-year-old method of preparing a blood smear on a glass slide, staining it, and examining it under a microscope to look for the parasite genus plasmodium. While several rapid diagnostic tests are also currently available, they still have shortcomings com-pared to microscopically analysis [2]. In the regions worst affected by malaria, reliable diagnoses are often difficult to obtain, and treatment is routinely pre- scribed based only on symptoms. Accurate diagnosis is clearly important, since false negatives can be fatal, and false positives lead to increased drug resistance, unnecessary economic burden, and possibly the failure to treat dis- eases with similar early symptoms such as meningitis or typhoid. The scale of the problem is huge: annually there are 300-500 million cases of acute malaria illness of which 1.1-2.7 million are fatal, most fatalities being among children under the age of five [3], [4], [5]. The lack of access to diagnosis in developing countries is largely due to a shortage of expertise, with a shortage of equipment being a secondary 1 * S M Raju is with Department of Computer Science, International Islamic University Malaysia, Gombak, Malaysia (e-mail: [email protected]). 2 Ali Mohammad Tarif is with Department of Computer Science, International Islamic University Malaysia, Gombak, Malaysia (e-mail: [email protected]). factor. For example, a recent survey carried out in Uganda [6] found 50% of rural health centers to have microscopes, but only 17% had laboratory technicians with the training necessary to use them for malaria diagnosis. Even where a microscopist is available, they are often oversubscribed and cannot spend long enough examining each sample to give a confident diagnosis. This situation has prompted an increasing interest in finding technological solutions to carrying out the diagnosis automatically with computer vision methods, taking advantage of existing equipment and compensating for the shortage of human expertise. Image processing and computer vision techniques can be used to identify parasites in blood smear images captured through a standard microscope. Given sufficient training data, the algorithms used in other medical imaging problems or computer vision tasks such as face detection can be applied to recognize plasmodia. Some studies have looked further at classifying the species and life cycle stage of parasites. Apart from the idea of using blood smear images captured directly from a microscope, there is a great deal of attention currently on other forms of point of care diagnosis for malaria. II. BACKGROUND In Africa, the estimated population at risk of malaria epidemics ranges from 52 to 144 million, according to methodologies used [16]. Populations affected by epidemics live in the highlands – or arid and semi-arid areas. Here, unusual rainfall and/or higher temperatures might play a strong role in triggering such epidemics, especially after an extended period of drought, thereby increasing general population vulnerability. Recent advances in research on malaria early warning systems are potentially useful to reduce the effects of epidemics. Here, different intervention decisions will be discussed within the context of research findings relevant to epidemic malaria. We use Convolutional Neural Network algorithm into this paper. Convolutional Neural Network (CNN) is contained at least one convolutional layer and after that followed by one or more fully connected layers as in a standard multilayer neural network. Spatial analysis models based on Geographic Information Systems (GIS) have been used to develop predictive algorithms for malaria vector distribution [17-19]. These, along with the 3 Mishkat Nur Rahman is with Department of Computer Science, International Islamic University Malaysia, Gombak, Malaysia (e-mail: [email protected]). 4 Md. Shariful Islam is with Department of Computer Science, International Islamic University Malaysia, Gombak, Malaysia (e-mail: [email protected]). 2 identification of the risk for human infection in epidemiological studies [20-21]. Convolutional Neural Network (CNN) was used in our project. In machine learning, a convolutional neural network (CNN, or ConvNet) is a class of deep, feed-forward artificial neural networks that has successfully been applied to analyzing visual imagery.5 CNNs use a variation of multilayer perceptron’s designed to require minimal preprocessing. They are also known as shift invariant or space invariant artificial neural networks (SIANN), based on their shared-weights architecture and translation invariance characteristics. Convolutional networks were inspired by biological processes in that the connectivity pattern between neurons resembles the organization of the animal visual cortex. Individual cortical neurons respond to stimuli only in a restricted region of the visual field known as the receptive field. The receptive fields of different neurons partially overlap such that they cover the entire visual field. III. RELEVANT WORKS A number of studies have looked at image processing and computer vision methodology for automated diagnosis of malaria from blood smears. In vision terms this is an object detection problem, and some previous work is reviewed in [12]. There has also been work in comparing these methods with other forms of diagnosis [8]. [10] uses neural networks with morphological features to identify red blood cells and possible parasites present on a microscopic slide. The results obtained with this technique were 85% recall and 81% precision using a set of 350 images containing 950 objects. In [11] a distance weighted k-nearest neighbor classifier was trained with features extracted by use of a Bayesian pixel classifier which was used to mark the stained pixels. The results achieved by this method were 74% recall and 88% precision. Color space and morphological heuristics were employed to segment red blood cells and parasites by using an optimal saturation threshold [9] using a set of 55 images. Multi-class parasite identification, attempting to classify the type and life cycle stage of detected parasites has also been attempted [13]. In [1], Quinn et al. propose a method to automatically dia


Download document

Buy ($6.60)


Document type





Artificial Intelligence, Health Informatics, Technology.



Related documents

Sentiment Analysis of Apple products using SVM and Maximum Entropy Model

Constructing A Network Layout

Heterosis studies in wheat