MÉTODO POINT OF CARE PARA O DIAGNÓSTICO DA SÍFILIS CONGÊNITA POR TÉCNICA ESPECTROSCÓPICA

Abstract

Syphilis is a curable sexually transmitted infection (STI) with accessible and inexpensive treatment, difficult to control, with a significant increase in incidence over the last decade. Infected pregnant women can transmit the disease to the fetus, characterizing congenital syphilis (CS), which may be asymptomatic. Fourier Transform Infrared Spectroscopy (FTIR) is a promising tool that can facilitate the detection and diagnosis of several diseases, which can improve early screening. However, the diagnostic potential of CS using this fast, non-invasive technology, especially associated with machine learning (ML), has not yet been investigated. The aim of the study was to develop a method based on FTIR and ML for the direct analysis of saliva in order to obtain an early diagnosis of CS in infants. Infants aged 0 to 12 months, of both sexes, born or admitted for treatment of CS at the Maria Aparecida Pedrossian University Hospital (HUMAP), Federal University of Mato Grosso do Sul (UFMS), Campo Grande - MS, participated in the study. For control purposes, healthy infants or those hospitalized for treatment of other pathologies, aged 0 to 12 months, participated. Sampling was convenience, and recruitment took place from 2023 to 2024. Spectra were obtained from 27 saliva samples of infants, 14 from the CS group and 13 from the control group. In order to balance the number of samples per class, an upsampling process was applied, totaling 20 samples for each group. Principal component analysis (PCA) and loanings evaluated sample variance, efficient discrimination, and the relevance for the variance observed between groups. The spectra were subjected to analysis and the results demonstrated differences in spectral patterns between the test and control groups. When applied to the quadratic optimized Support Vector Machine (Quadratic SVM) algorithm and the leave-one-out cross-validation (LOOCV) technique, the model achieved 90% accuracy, 100% sensitivity and 80% specificity, showing the potential for a screening test. Descriptors: