Classificação das fases do plasmodium em imagens do exame de gota espessa para a malária

Abstract

Malaria is a potentially fatal disease that spreads to humans through the bite of female Anopheles mosquitoes infected with Plasmodium parasite species. Prompt diagnosis of malaria is recommended by the WHO for all patients with suspected malaria before they receive treatment. Blood smear microscopy is one of the recommended diagnostic tests. However, blood smear microscopy requires experience, is time-consuming and is subject to intra- and inter-microscopic variability. Automated microscopy has the potential to overcome these problems. In this context, some computational methods based on machine learning for object detection and classification have been developed. This work proposes the detection and identification of Plasmodium stages in brightfield microscopy images of blood smears. Nine deep network models were analyzed through an ablation study in a 2-stage process: the first performing a binary classification and the second classifying the parasites detected in the first stage into one of its 4 phases. The models were tested with a dataset of public domain images. The best results were obtained with the Efficient Net BX deep network. With this network, an accuracy of 89,91% and an F1-score of 83,88% were obtained, which surpassed the results presented in previous studies.