Determination of Optical Properties of Skin Tissues Using Spatial Domain Frequency Imaging and Random Forests

The evaluation of optical properties of biological tissues has been pointed as an important tool for detection and diagnosis of tissue alterations. The Spatial Frequency Domain Imaging (SFDI) provides quantitative information about light absorption and scattering properties in tissues from measurements of diffuse reflectance. This technique requires the proper correlation between the measured values of diffuse reflectance of light by the tissue, Rd, at different spatial frequencies and the corresponding pair of absorption and reduced scattering coefficients \({\mu }_{a}\) and \({\mu {\prime}}_{s}\) , respectively. In this work, the machine learning technique of Random Forests was applied to provide a regression model that efficiently computes \({\mu }_{a}\) and \({\mu {\prime}}_{s}\) from Rd values. The database employed consisted of training and testing values of Rd at different spatial frequencies for different combinations of \({\mu }_{a}\) and \({\mu {\prime}}_{s}\) , obtained from Monte Carlo simulations. Results showed that the correlation coefficient R2 between predicted and expected values from the test group were 0.96 and 0.97, for \({\mu }_{a}\) and \({\mu {\prime}}_{s}\) , respectively. The relative average errors for each coefficient were, respectively, 1% and 0.004%, with standard deviations of 11% and 7%. These results point to the good accuracy and precision of the developed models. These models were applied to an in vivo study, where values of Rd from the dorsal region of the hand of a volunteer were obtained with SFDI equipment using light wavelength of 650 nm. The obtained images of \({\mu }_{a}\) and \({\mu {\prime}}_{s}\) showed enhanced contrast of blood vessels, pointing to the potential of the technique to identify vascular tissue alterations.