Aprendizado de máquina aplicado à elipsometria espectroscópica para o diagnóstico sorológico de malária

Abstract

Malaria is a serious disease that primarily affects tropical and less developed regions of the world. Infected patients need to be diagnosed and treated quickly to prevent the progression of the disease to potentially fatal stages. However, various technical and economic barriers hinder access to and adoption of currently available diagnostic methods. This study presents a new malaria diagnostic method that is simple, effective, low-cost, rapid, and does not require the use of biomarkers. To achieve this, an artificial intelligence-based code was developed, capable of differentiating healthy individuals from those reactive to malaria by processing Mueller Matrix spectroscopy data from blood serum samples. The developed models demonstrated performance with AUC values above 0.9 in classifying negative and malaria-reactive individuals, in addition to providing physically consistent results in the near-infrared region. Furthermore, the models can distinguish patients from endemic and non-endemic areas, offering a promising approach for diagnostics in different epidemiological contexts.