Microscopy image analysis using deep metric learning: a study on the detection and classification of leishmania amastigotes and other parasites

Abstract

Visceral Leishmaniasis, a severe form of the disease caused by the parasite Leishmania donovani, is fatal in over 95% of untreated cases and primarily affects low-income individuals with limited access to healthcare. The standard diagnosis involves identifying the parasite’s amastigotes, which are small and difficult to find, making the examination a challenging task that requires skill. To assist healthcare professionals, this study proposes a new approach that combines deep metric learning with supervised classification for the rapid detection of visceral leishmaniasis. The methodology divides images into small fragments (patches) to enhance the evaluation of four loss functions, which help a Support Vector Machine (SVM) diagnose the disease. The results showed that the Circle loss function performed best, with 98.3% sensitivity and 99.3% specificity. In addition to leishmaniasis, we explored the performance on other parasitic infections, such as Babesia, Toxoplasma, Trypanosoma, Plasmodium, and Schistosoma, which also demonstrated impressive results, with high precision and sensitivity. This approach suggests that artificial intelligence can be a valuable tool for improving the diagnosis of neglected tropical diseases, making it more accessible and efficient.