Performance Analysis of Various Feature Sets for Malaria-Infected Erythrocyte Detection

Abstract

Malaria being prevalent disease in urban areas, demands its accurate and fast diagnosis. Due to malaria infection in human being, the erythrocyte features got distorted. To diagnose these, various techniques have been developed, i.e., machine learning-based system, rapid diagnostic test, quantitative buffy coat, etc. In machine learning, the system performance depends on the feature set and classifier model. In this paper, the analysis of the importance of the feature set on malaria-infected erythrocyte classification has been performed. Further, a classifier model based on ANN-GA has been developed to classify the erythrocyte. The process consists of illumination correction, erythrocyte segmentation, feature extraction with or without feature selection techniques, and classification. Erythrocytes segmentation is done using image binarization with marker-controlled watershed segmentation. The six feature sets (morphological feature, texture and intensity feature) have been evaluated using various classifiers such as support vector machine (SVM), k-nearest neighbor (k-NN), and Naive Bayes to choose the better feature set. From the experimental results, it has been observed that the feature set $\textit{f}_6$ (combination of morphological, texture and intensity feature ranked with ANOVA) outperforms other feature sets. Further, erythrocyte classification has been performed using ANN-GA with $\textit{f}_6$ feature set. It may also conclude that the various features such as morphological feature, texture and intensity feature are equally important to detect the malaria-infected erythrocyte.