Fig. 1. Overview of dental optical properties computation and measurement technique

Fig. 2. Optical transmission processes in teeth. (a) Natural tooth; (b) restorative tooth

Fig. 3. Optical transmission model of bilayer structure for tooth

Fig. 4. Structure of the OPTNet model

Fig. 5. Structure of spatial attention mechanism module

Fig. 6. Comparison of diffuse reflectance under different optical properties. (a) Diffuse reflectances under different reduced scattering coefficients of the dentin layer with reduced scattering coefficient of the ceramic layer of 11.0 mm^-1, absorption coefficient of the ceramic layer of 0.06 mm^-1, and absorption coefficient of the dentin layer of 0.1 mm^-1; (b) diffuse reflectances under different absorption coefficients of the dentin layer with reduced scattering coefficient of the ceramic layer of 11.0 mm^-1, absorption coefficient of the ceramic layer of 0.06 mm^-1, and reduced scattering coefficient of the dentin layer of 15.0 mm^-1

Fig. 7. Dental optical properties acquisition scheme. (a) Optical path diagram; (b) optical acquisition system; (c) laser; (d) acquired image

Fig. 8. Measurement of light spot distribution on the tooth surface and the relative diffuse reflectivity. (a) Physical image of tooth; (b) measured light spot; (c) relative diffuse reflectance distribution with reduced scattering and absorption coefficients of the ceramic layer of 18.802 mm^-1 and 2.763 mm^-1, and reduced scattering and absorption coefficients of the dentin layer of 10.886 mm^-1 and 0.481 mm^-1

Fig. 9. Comparison of predicted and true optical properties of dentin layer on real samples. (a)Reduced scattering coefficient; (b)absorption coefficient

Fig. 10. MRE obtained through inversion on the biased sample validation sets DS 1, DS 2, and DS 3

Fig. 11. Comparison of optical properties inversion results for biased samples. Inversion results of reduced scattering coefficient by (a) FNN, (b) CNN, (c) CNN-FCN, and (d) proposed algorithm; inversion results of absorption coefficient by (e) FNN, (f) CNN, (g) CNN-FCN, and (h) proposed algorithm