This research was conducted under the supervision of Dr. François Rivest as part of the requirements for CISC-867 Deep Learning course at Queen's University.
Malaria is a serious global health problem of humans caused by the Plasmodium parasites transmitted through the bite of female Anopheles mosquito. Visual examination of blood smears under the microscope by expert technicians is the standard way of diagnosing malaria parasite infected red blood cells. This method is inefficient since the diagnostic accuracy heavily depends on the experience and the expertise of the person doing the examination. Although automatic image recognition techniques based on machine learning have been applied toward the task of malaria diagnosis of blood smears before, the delivered performance has not been sufficient so far. In this study we propose two deep learning models based on convolutional neural networks (CNNs) toward the task of automated classification of malaria infected and uninfected thin blood smear images. Using a total number of 27598 images from human blood cells our 21- and 16- layer CNNs could achieve average accuracy of 96.08% and 95.53% respectively. Comparing our CNN-based solutions with transfer learning models, we can observe superior performance of CNN models over transfer learning method along with other traditional classifiers.