Medical use of deep learning: Malaria testing using pre-trained ResNet

Abstract

Malaria is a contagious blood disease caused by Plasmodium parasites transmitted by the bite of an anopheles female mosquito, or in some case by humans. Malaria Test can be done ether using microscope or by Rapid Test. The first method is very accurate but time-consuming because it needs the intervention of human. On the other hand the second one is very fast but not accurate. Several works have contributed to this subject to combine the two criteria using Artificial Intelligence techniques. Our solution is based on the Deep Learning, it uses a Residual Neural Network Model applied to labeled images taken by a microscope and classified by the healthcare experts (Supervised Learning), the data set used is proposed by NIH (National Institutes of Health). In this study, we evaluate the performance of the ResNet model as a classifier. The validation of the results demonstrates the compromises: model size, training time and accuracy. We have used our model after, to build a desktop GUI application and a Web Service architecture that will automate the Malaria test process and help healthcare agents, doctors or normal person for a final goal of saving thousands of lives.