Enhanced Detection of Leishmania Parasites in Microscopic Images Using Machine Learning Models

Abstract

Cutaneous leishmaniasis is a parasitic disease that poses significant diagnostic challenges due to the variability of results and reliance on operator expertise. This study addresses the development of a system based on machine learning algorithms to detect Leishmania spp. parasite in direct smear microscopy images, contributing to the diagnosis of cutaneous leishmaniasis. Starting with acquiring and labeling 500 images, an experimental design was implemented, including preprocessing and segmentation techniques such as Otsu, local thresholding, and Iterative Global Minimum Search (IGMS) to improve parasite detection. The phenotypic features of the parasites were extracted, focusing on morphology, texture, and color. Machine learning models (ANN, SVM, and RF) optimized through Grid Search were applied for classification. The model with the best results was a Support Vector Machine (SVM), achieving a sensitivity of 91.87% and a specificity of 89.21% at the crop level. Compared with previous studies, these results highlight the relevance and consistency of the methodology used, supporting the initial hypothesis. This suggests that machine learning techniques offer a promising path toward improving the diagnosis of cutaneous leishmaniasis.