Erythrocyte Classification using Multi-Layer Perceptron, Naïve Bayes Classifier, RBF Network and SVM
Dyah Aruming Tyas1, Sri Hartati2, Agus Harjoko3, Tri Ratnaningsih4
1Dyah Aruming Tyas, Department of Computer Science and Electronics, Faculty of Mathematics and Natural Sciences, Universitas Gadjah Mada, Yogyakarta, Indonesia.
2Sri Hartati*, Departmenttof Computerr Scienceeandd Electronics, Facultyyof Mathematicssand NaturalSSciences, Universitass Gadjahh Mada, Yogyakarta, Indonesia.
3Agus Harjoko, Departmentt ofCComputer Scienceeand Electronics, Facultyyoff Mathematicssandd Natural Sciences, Universitas Gadjah Mada, Yogyakarta,, Indonesia.
4Tri Ratnaningsih, Department of Clinical Pathology, Faculty of Medicine, Public Health, and Nursing, Universitas Gadjah Mada, Yogyakarta, Indonesia.
Manuscript received on November 22, 2019. | Revised Manuscript received on December 15, 2019. | Manuscript published on December 30, 2019. | PP: 2024-2028 | Volume-9 Issue-2, December, 2019. | Retrieval Number: B3231129219/2019©BEIESP | DOI: 10.35940/ijeat.B3231.129219
Open Access | Ethics and Policies | Cite | Mendeley
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Abstract: Several diseases can be diagnosed based on the appearance of abnormal erythrocytes, among others anaemia and thalassemia. Process of examination peripheral blood smear manually is time-consuming and subjective. Currently, the process of examination peripheral blood smear by laboratory assistants can be assisted with digital image processing technology so that it can speed up the examination time and avoid subjectivity. This research begins with the process of microscopic image acquisition, then preprocessing, segmentation, feature extraction and classification. The microscopic image acquisition is carried out using an additional special camera on a microscope. In this study, we used peripheral blood smear of thalassemia patients and healthy individuals. We convert the RGB image to grayscale image and perform the median filtering in the preprocessing stage. In the segmentation stage, we used the watershed distance transform method. As a segmentation result, we got 7108 erythrocyte images consisting of nine types of erythrocytes. In feature extraction, we used shape, color and texture characteristics to represent erythrocytes. The combination of these three features is used as classifier’s input. One crucial stage in digital image processing technology is object classification. In this study, erythrocyte classification is done by comparing four types of the classifier to determine the best classifier performance in this case. Multi-Layer Perceptron (MLP), Naïve Bayes classifier, RBF Network, and SVM used as classifiers in this study. Experimental results showed that MLP got the highest performance with 89.6% accuracy, 89.3% precision and 89.6% recall. Furthermore SVM came in second place, followed by RBF Network and Naïve Bayes classifier.
Keywords: Classification, erythrocyte, MLP, Naïve Bayes classifier, RBF Network, SVM.