Identifying Malaria Infection in Red Blood Cells using Optimized Step-Increase Convolutional Neural Network Model

Abstract

A vast number of image processing and neural network approaches are currently being utilized in the analysis of various medical conditions. Malaria is a disease which can be diagnosed by examining blood smears. But when it is examined manually by the microscopist, the accuracy of diagnosis can be error-prone because it depends upon the quality of the smear and the expertise of microscopist in examining the smears. Among the various machine learning techniques, convolutional neural networks (CNN) promise relatively higher accuracy. We propose an Optimized Step-Increase CNN (OSICNN) model to classify red blood cell images taken from thin blood smear samples into infected and non-infected with the malaria parasite. The proposed OSICNN model consists of four convolutional layers and is showing comparable results when compared with other state of the art models. The accuracy of identifying parasite in RBC has been found to be 98.3% with the proposed model.