Chagas disease remains a major neglected tropical disease, particularly during the chronic phase, where diagnosis is challenged by low parasitemia and the extensive genetic and antigenic diversity of Trypanosoma cruzi. Current serological approaches require multiple antigen-dependent assays, increasing cost, complexity, and turnaround time, especially in resource-limited settings. Here, we report a label-free diagnostic strategy combining attenuated total reflectance Fourier-transform infrared (ATR-FTIR) spectroscopy of human serum with machine learning to discriminate individuals with chronic Chagas disease from noninfected controls. Serum samples (n = 68; 34 positive and 34 negative, previously confirmed by serology and PCR) were analyzed in the 1800–900 cm–1 spectral range. Spectral data were denoised using Savitzky–Golay smoothing and fast Fourier transform filtering, normalized by a modified standard normal variate method, and reduced by principal component analysis (PCA). Support vector machine (SVM) classifiers trained on PCA scores achieved optimal performance using a quadratic kernel (C = 1). External validation using an independent test set yielded 100% sensitivity, 91% specificity, and 95.5% accuracy, with confidence intervals assessed by Wilson score statistics. Spectral loadings indicate that protein-related vibrational modes, particularly amide I and II bands, dominate class separation, consistent with alterations in humoral immune components during chronic infection. These results demonstrate that ATR-FTIR spectroscopy coupled with machine learning provides a rapid, scalable, and antigen-independent diagnostic approach for chronic Chagas disease. This methodology holds strong potential for point-of-care screening and large-scale epidemiological surveillance, particularly in settings where conventional laboratory infrastructure is limited.
Maranni et al. (2026) studied this question.