Malaria parasite detection using deep learning algorithms based on (CNNs) technique

Abstract

Malaria is a life-threatening disease caused by female anopheles mosquito bites that are prevalent in many regions of the world. We introduce a deep convolutional neural network (CNN) to improve malaria diagnosis accuracy using patches segmented from microscopic images of red blood cell smears. We design the automatic parasite detection in blood from Giemsa-stained smears using three CNN pre-trained models such as VGG19, ResNet50, and MobileNetV2. As the CNNs are poorly performing for small datasets, we introduce the transfer learning technique. Transfer learning involves acquiring visual features from large general datasets and resolving issues using small datasets. We use a transfer learning approach to detect and classify malaria parasites with three CNN pre-trained models. We evaluated proposed CNN models experimentally using the National Institute of Health (NIH) Malaria Dataset. Our proposed model achieves an accuracy of almost 100%.