An integrated fiber-optic probe combined with support vector regression for fast estimation of optical properties of turbid media

• A novel integrated fiber probe is proposed to estimate the optical properties.
• SVR models the relationship between diffuse reflectance and optical properties.
• This approach is fast, efficient and do not rely on the simulation results from MC.
• This approach avoids the theoretical limitation from RTE and DT.
• SVR overcomes the over-fitting problem and fits the non-linear relationship well.

A fiber-optic probe system was developed to estimate the optical properties of turbid media based on spatially resolved diffuse reflectance. Because of the limitations in numerical calculation of radiative transfer equation (RTE), diffusion approximation (DA) and Monte Carlo simulations (MC), support vector regression (SVR) was introduced to model the relationship between diffuse reflectance values and optical properties. The SVR models of four collection fibers were trained by phantoms in calibration set with a wide range of optical properties which represented products of different applications, then the optical properties of phantoms in prediction set were predicted after an optimal searching on SVR models. The results indicated that the SVR model was capable of describing the relationship with little deviation in forward validation. The correlation coefficient (R  ) of reduced scattering coefficient μs′ and absorption coefficient μa   in the prediction set were 0.9907 and 0.9980, respectively. The root mean square errors of prediction (RMSEP) of μs′ and μa in inverse validation were 0.411 cm−1 and 0.338 cm−1, respectively. The results indicated that the integrated fiber-optic probe system combined with SVR model were suitable for fast and accurate estimation of optical properties of turbid media based on spatially resolved diffuse reflectance.

Figure optionsDownload as PowerPoint slide