A Fully Automated System to Detect Malaria Parasites and their Stages from the Blood Smear

Abstract

Malaria is a common disease caused by female Anopheles mosquitoes. There are mainly four species of malaria and each species has four stages. Early detection of malaria is a key factor in proper patient management as well as reducing death from malaria. The traditional way of detecting malaria is time-consuming and varies depending on the expertise of the pathologist. To avoid the person dependency for appropriate detection several studies implemented an automated method for detecting malaria parasites. The automatic detection of malaria parasites with their stages from blood smear has been proposed in this work. The blood smear images of malaria patients were collected from the online database. After applying some preprocessing U-Net was used to segment the RBC (Red Blood Cell) from blood smear images, CNN to identify infected RBC by malaria parasites, and finally, an award-winning neural network called VGG16 was used to recognize the different types and stages of malaria. The segmentation accuracy and specificity of the U-Net model were 97.67% and 92.05% respectively. The detection accuracy and specificity of infected RBC were 100% and 95% by using the CNN model. The average accuracy and specificity of the VGG16 model for malaria species detection 95.55% and 94.75% respectively. The average staging accuracy and specificity of different types of malaria parasites for the ring stage were respectively 96.25% and 94.82% by applying VGG16.