FTIR-based machine learning for prediction of malignant transformation in oral epithelial dysplasia.

Oral squamous cell carcinoma (OSCC) is an aggressive cancer with a poor prognosis. Oral epithelial dysplasia (OED) is a precancerous lesion associated with an increased risk of malignant transformation (MT) into OSCC. However, current histopathological methods for diagnosing OED are subjective and ineffective in assessing MT risk. FTIR provides a comprehensive biochemical profile of tissues, known as "biomolecular fingerprinting". Previously we developed an FTIR-based OSCC-Benign classifier that accurately distinguishes OSCC from benign tissue. In this study, we evaluated whether this classifier could also predict MT risk in OED. Thirty OED patient biopsies with documented MT outcomes were analyzed, including 12 with and 18 without MT. FTIR images were acquired from six regions of interest (ROIs) per tissue section, yielding an average epithelial spectrum for each ROI, and a total of 180 spectra for model evaluation. The OSCC-Benign classifier achieved an accuracy of 81.7% with an F1 score of 0.77 at the ROI level, and an accuracy of 83.3% with an F1 score of 0.8 at the biopsy level in predicting MT in OED. Our findings suggest that OEDs with biomolecular fingerprints similar to OSCC carry a higher risk of MT, while those resembling benign tissue carry a lower risk, providing new insight into the malignant transformation process. In summary, the FTIR-based machine learning approach outperforms traditional histopathology in predicting MT risk in OED, potentially offering a quantitative and objective tool for clinical diagnosis.