AI-driven infrared spectroscopy in salivary oral leukoplakia

Objectives Oral potentially malignant disorders (OPMDs), particularly leukoplakia, are strongly linked to an increased risk of oral squamous cell carcinoma. Saliva has gained attention as a diagnostic biofluid due to its simple, non-invasive collection and complex biochemical composition. This study aimed to investigate the application of Fourier-transform infrared (FTIR) spectroscopy combined with machine learning to identify salivary biochemical changes. Materials and methods Saliva samples were collected from healthy control participants ( n = 22), participants with oral leukoplakia ( n = 22). Spectral data were acquired by ATR-FTIR, pre-processed with derivative analysis, normalization, and baseline correction, then analyzed using PCA and classified with Extra Trees, Random Forest, and XGBoost algorithms. Results The analysis revealed consistent biochemical differences within the 900–1800 cm⁻¹ region, mainly associated with proteins (Amide I and II), nucleic acids, and collagen. PCA revealed partial but relevant group separation, while machine learning models, especially Extra Trees and Random Forest, the MLP classifier achieved an AUC of 0.992 for initial dysplasia and 0.974 for severe dysplasia versus controls, indicating promising classification perfomance. Conclusion These findings support FTIR spectroscopy integrated with machine learning, as a powerful, non-invasive approach for salivary profiling in leukoplakia. This optical biopsy method shows promise as an auxiliary diagnostic tool for early detection and monitoring of OPMDs, enhancing preventive strategies in oral oncology.