Visceral Leishmaniasis, a severe type caused by the Leishmania donovani parasite complex, is fatal in over 95% of untreated cases and predominantly affects the poor and vulnerable with limited healthcare access. Parasitological processes are the gold standard for diagnosing VL; they entail direct microscopic inspection of amastigotes about 2–4 μm in diameter, which can quickly become a time-consuming, exhausting task and require an expert skill level.
Aiming to assist physicians, this study proposes an alternative approach combining deep metric learning with supervised classification for the rapid and reliable detection of human visceral leishmaniasis. The suggested methodology segments images into patches for discernability during the evaluation of four deep metric learning loss functions to extract features, which are utilized by a Support Vector Machine (SVM) for the diagnosis of visceral leishmaniasis.
This process was thoroughly assessed using key metrics like the Matthew Correlation Coefficient (MCC), sensitivity, and specificity, which revealed that Circle loss outperforms other losses with 98.3% sensitivity, 99.3% specificity, and 97.7% MCC. Overall, all of the functions evaluated performed well in quantitative assessments, implying that AI’s application to medical diagnostics offers considerable benefits, particularly in cost-effectively assisting physicians in rapidly and accurately detecting neglected tropical diseases.
This is an undergraduate thesis.