A machine-learning model for quantitative characterization of human skin using photothermal radiometry and diffuse reflectance spectroscopy

摘要

We have recently introduced a novel methodology for noninvasive assessment of structure and composition of human skinin vivo. The approach combines pulsed photothermal radiometry (PPTR), involving time-resolved measurements of midinfraredemission after irradiation with a millisecond light pulse, and diffuse reflectance spectroscopy (DRS) in visiblepart of the spectrum (400–600 nm). The experimental data are fitted simultaneously with respective predictions from afour-layer Monte Carlo (MC) model of light transport in human skin.The described approach allows assessment of the contents of specific chromophores (melanin, oxy-, and deoxyhemoglobin),as well as scattering properties and thicknesses of the epidermis and dermis. However, the involved multidimensionaloptimization with a numerical forward model (i.e., inverse MC) is computationally very expensive. Inaddition, each optimization task is repeated several times to control the inevitable numerical noise and facilitate escapefrom local minima. Thus, assessment of 14 free parameters from each radiometric transient and DRS spectrum takesseveral hours despite massive parallelization using CUDA technology and a high-performance graphics card.To alleviate this limitation, we have constructed a computationally very efficient predictive model (PM) based on machinelearning. The PM is an ensemble of decision trees (random forest), trained using ~12,000 "pairs" of various skin parametercombinations and the corresponding PPTR signals and DRS spectra, computed using our forward MC model. We analyzethe performance of such a PM by means of cross-validation and comparison with the inverse MC approach.