Diagnóstico diferencial da esquistossomose por espectroscopia ATR-FTIR integrada a algoritmos de inteligência artificial

Abstract

Schistosomiasis is classified as a Neglected Tropical Disease and represents an important public health problem in Brazil, with Schistosoma mansoni as the etiological agent in South America. The currently available diagnostic methods present limitations in accuracy, sensitivity, and specificity, in addition to requiring technical expertise both in sample processing and in result interpretation. In this study, Fourier Transform Infrared Spectroscopy with Attenuated Total Reflectance (ATR-FTIR) was employed to identify spectral signatures capable of assisting in the differential diagnosis of schistosomiasis. Serum samples were obtained from the repository of the Laboratory of Immunobiology and Parasite Control at the Federal University of Minas Gerais. For ATR-FTIR analysis, 54 serum samples from individuals positive for schistosomiasis and 58 from non-infected individuals living in endemic and non-endemic areas were used. All schistosomiasis-positive participants presented exclusive infection by Schistosoma mansoni, confirmed through complementary parasitological examinations, ensuring the specificity of the analyses. Artificial intelligence techniques, focusing on machine learning methods, were applied for spectral classification. The results indicated that preprocessing with the first derivative provided the best model performance, resulting in more efficient diagnostic classification. Using the Random Forest algorithm, accuracy, sensitivity, specificity, area under the curve, and F1-score values of 80%, 81%, 79%, 0.83, and 0.80, respectively, were obtained. SHAP features extracted from the infrared spectra highlighted the main vibrational modes associated with proteins, nucleic acids, and carbohydrates in serum, responsible for discriminating between individuals infected and non-infected with S. mansoni. Therefore, the use of ATR-FTIR combined with machine learning proves to be a promising screening method for endemic areas, particularly due to its non-invasive, highly sensitive, and reagent-free nature.