Automated Detection of Malaria Parasites Using Image Processing and Convolutional Neural Network Classifier.

Abstract

Malaria, being one of the most dangerous parasite diseases endangering human life and leads to high mortality and morbidity rates, affects millions of people, especially in subtropical and tropical regions. Because of the dependence on human skills and the inaccuracy of manual analysis, conventional diagnostic techniques, like blood smear testing, which could be carried out under a microscope, confront numerous difficulties. Thus, automating malaria detection using deep learning (DL) or machine learning (ML) algorithms presents viable ways for decreasing diagnosis time, increasing scalability, and improving accuracy. Since P. vivax and Plasmodium falciparum lead to the majority of deaths and severe cases, multi-class convolutional neural network (CNN)-based model has been developed in the presented work for classifying cells from the blood smears that are infected with P. vivax and P. falciparum and uninfected cells. Instead of utilizing transfer learning from the previously trained models, this is accomplished through creating as well as training CNN from the ground up. Kaggle data-set, which includes 27,558 images of both uninfected and infected people, has been used for training and testing the suggested network. The images have been separated to 13,779 images of uninfected persons, 6,889 images of people who had P. vivax malaria and 6,890 images of people who had P. falciparum malaria. A number of preprocessing techniques, such as denoising, blurring, and morphological processing, have been applied to the images. With a 95% sensitivity rate, a 96.50% accuracy rate, a 96.5% F1-score rate, and a 97.60% specificity rate, the suggested model outperformed other DL algorithms in the evaluation accuracy. The findings show how well the suggested model works as a tool for assisting clinicians diagnose malaria by lowering the need for manual analysis.