AVALIAÇÃO DA ESPECTROSCOPIA UV-VIS ASSOCIADA AO APRENDIZADO DE MÁQUINA PARA DIAGNÓSTICO DE LEISHMANIA INFANTUM EM CÃES

Abstract

Abstract Canine Visceral Leishmaniasis (CVL) is a severe disease caused by the protozoan Leishmania infantum, transmitted by arthropods of the genus Lutzomyia. It is a zoonosis of major importance, and affected animals may present clinical signs such as weight loss, weakness, intermittent fever, and renal and hepatic impairment. The definitive diagnosis of leishmaniasis remains imprecise, since it shares similarities with other diseases. The diagnostic techniques currently recommended by the Ministry of Health of Brazil, for dogs, are the immunochromatographic rapid test and the enzyme-linked immunosorbent assay (ELISA), which, although widely used, still present issues related to sensitivity and specificity. Currently, optical spectroscopy techniques combined with machine learning-based analyses have been investigated for the diagnosis of infections caused by various pathogens, including L. infantum. Spectroscopy is an accessible and low-cost methodology, and has shown promising results in differentiating positive and negative samples for Brucella, Leishmania, among others. However, to date, most studies have relied on dichotomous comparisons (infected vs. non-infected), leaving uncertainties regarding the actual ability of the methodology to determine a specific infection. In this context, the present study aimed to evaluate in greater detail the serological diagnostic methodology of L. infantum based on UV-Vis spectroscopy and machine learning, not only in terms of its ability to differentiate positive and negative samples for L. infantum, but also to assess co-infections and samples positive for other common infectious agents in dogs, such as Ehrlichia canis. The results obtained indicate that co-infection with Ehrlichia hampered the accurate identification of animals infected and non-infected with Leishmania (accuracy below 50%). Nevertheless, good discriminatory capacity was observed in distinguishing positive and negative samples for E. canis (accuracy of 77.8%). Expanding the dataset and implementing parameter adjustments and machine learning methods will be necessary to improve the discriminatory ability of the methodology and make it feasible as a diagnostic tool.

Keywords: Leishmaniasis, spectroscopy, diagnosis.