Spatial Chemical Imaging and Quantitative Prediction of Dentin Demineralization via Raman Spectroscopy

This study develops a non-destructive analytical framework based on Raman spectroscopy for spatial chemical imaging and quantitative assessment of dentin demineralization. Human dentin samples treated with gradient EDTA and natural caries samples were examined. SEM−EDS was used to analyze surface morphology and elemental (Ca, P) contents. Single-point Raman spectra and Raman mapping datasets were then collected. Raman spectroscopy captured coordinated changes in mineral-related bands, collagen-associated bands, and mineral-to-matrix ratios, while MCR-ALS unmixed Raman mapping data into spatially resolved mineral-rich and organic/mixed components. Four PLSR modelling strategies were employed to build predictive models, including full-spectrum PLSR, full-spectrum FiPLS, manual 12-feature PLSR, and manual 12-feature RFE-PLSR. The manual-12-RFE-PLSR model achieved the highest predictive performance for Ca (R2 = 0.895, RMSE = 2.508) and P (R2 = 0.872, RMSE = 1.217), with comparable performance from the interval-selection FiPLS model. Importantly, manual-12-RFE-PLSR and full-spectrum FiPLS maintained acceptable accuracy (R2 > 0.75) when generalized to highly heterogeneous natural caries. This framework enables quantitative elemental tracking and spatial mapping of component distributions, demonstrating potential for providing real-time, objective decision support in minimally invasive dentistry.