Deep Learning for Detecting Malaria Parasites of Infected Red Blood Cells in Thin Blood Smear Images

Abstract

Malaria is a significant global health issue, with 241 million people infected and resulting in 627,000 deaths in 2020, officially reported by the World Health organization (WHO). In addition, during the Covid-19 pandemic, 47,000 people died because of a reluctance to receive treatment. In Thailand, Malaria still spreads in distant communities where restrictions are in place for military deployments due to the high risk of infection. Therefore, the 8,000 or so military personnel who deploy on missions close to the country’s borders are actively monitored by the Armed Forces Research Institute of Medical Sciences (AFRIMS). The lack of medical personnel in these remote settlements, however, slows detection and adversely impacts the health and lives of military soldiers working in these locations. Because of their comparative effectiveness to traditional learning algorithms, deep learning technologies are used as a tool for medical screenings. In this study, the YOLOv3 and the DenseNetl21 are used to diagnose malaria infection using thin film blood smears. The results show that testing on normal slide datasets can distinguish between normal red blood cells and malaria-infected red blood cells in four species, including Falciparum, Vivax, Malariae, and Ovale, with accuracy for infection classification at 98.08%, sensitivity at 98.05%, and specificity at 99.73%. Furthermore, when the hard slide dataset is examined, the infection classification’s accuracy, sensitivity, and specificity are 98.48%, 90%, and 99.24%, respectively. In normal slide datasets, this detection method yields a positive hit rate for malaria-infected red blood cells and normal red blood cells of 98.05% for the former and 92.65% for the latter.