Plasmodium Life Cycle-Stage Classification on Thick Blood Smear Microscopy Images using Deep Learning: A Contribution to Malaria Diagnosis
F. A. S. Araujo, Nathalia D. Colares, U. P. Carvalho, Cícero Ferreira Fernandes Costa Filho, M. G. F. Costa · 2023
Malaria is a life-threatening disease spread to humans through the bites of female Anopheles mosquitoes infected with Plasmodium parasite species. Prompt malaria diagnosis is recommended by WHO for all patients with suspected malaria before they are given treatment. The blood smear microscopy is one of the recommended diagnostic tests. However, blood smear microscopy requires expertise, is time consuming and subject to intra and inter microscopist variability. Automated microscopy has the potential to overcome the problems mentioned. In this context, some computational methods based on machine learning for object detection and classification have been developed aiming to automatic parasitological diagnosis of malaria. This work proposes the classification of plasmodium life cycle-stage in patches of microscopy images of blood smears. We analyzed three deep network models through an ablation study in a 2-level classification: the first comprising a binary classification where the four Plasmodium Life Cycle-Stages (ring, trophozoite, schizont and gametocyte) are considered as a single class, of infected red blood cells, At the second level the infected cells are classified according to the Plasmodium Life Cycle-Stage into four classes. The models were tested with a public domain image dataset. The best results were obtained with the Efficient Net B7 deep network. With this network an accuracy of 87.95% and an Fl-score of87.80% were obtained, which outperformed the results presented in previous studies.