Fiber-Optic Raman Sensor for Early Dental Caries Detection: Performance Evaluation and Robustness to Probe Positioning

Early detection of dental caries remains a significant clinical challenge, as conventional diagnostic methods lack sensitivity for incipient lesions. Raman spectroscopy offers high chemical specificity for enamel characterization; however, clinical translation is hindered by the complexity of conventional polarized confocal systems. In this work, we present a Raman-based fiber-optic sensing approach for the detection and classification of dental enamel conditions, including sound, affected, and carious tissues. A custom fiber-optic probe was developed for remote measurements and evaluated against a reference polarized confocal Raman system. In addition to spectral discrimination, key factors affecting sensing performance were investigated, including spatial variability across enamel surfaces and angular sensitivity due to probe misalignment. Raman-derived features (carbonate-to-phosphate ratio, phosphate peak intensity, position, and bandwidth) were analyzed using a multinomial logistic regression classifier. The fiber-optic sensor achieved an overall classification accuracy of 73% (F1-scores: 0.55 sound, 0.63 affected, 0.9 carious), confirmed by leave-one-tooth-out cross-validation. Probe misalignment studies revealed robustness up to 10° angular deviation. These results demonstrate that a simplified non-polarized fiber-optic Raman system provides competitive diagnostic performance and clinically relevant robustness, supporting its development as a point-of-care sensing platform for early dental caries detection.