Avaliação de métodos clássicos de detecção de características no auxílio à identificação de amastigotas de leishmaniose

Abstract

Visceral leishmaniasis (VL) is an anthropozoonosis, of high lethality, transmited by the popularly known sand flies. Currently, the diagnosis of this disease is complex due to the similarity of its symptoms with other diseases, the gold standard of diagnosis being the parasitological examination that deals with the visual analysis of the parasite leishmania in biopsy material, this analysis being subject to errors due to factors such as the minuscule size of the parasite to be observed (about 2–4 µm in diameter) and the medical expertise in diagnosing this disease. In this study, from a patch-based method, we evaluated several texture descriptors and classification algorithms that can help identify these parasites. More precisely, a set of fragments (patches), extracted from microscopic images of bone marrow aspirates, were submitted to pre-processing, feature extraction and classification algorithms. The algorithms SIFT, ORB and Haralick were employed in the feature extraction, considering their detection, description and texture analysis. The classifiers SVM and KNN were used to classify the images, identifying the presence or absence of leishmania. The results obtained demonstrated the effectiveness of these techniques in the diagnosis of VL, with satisfactory accuracy in the identification of parasites. Therefore, this study highlights the potential of image processing techniques and machine learning algorithms as auxiliary tools in the accurate and rapid diagnosis of this serious disease, enabling effective treatment and better outcomes for affected patients.