Hyperspectral Imaging Combined With Machine Learning Methods to Quantify the Facial Skin Melanin and Erythema.

Abstract

Melanin deposition and erythema mainly constitute physiological responses of the skin to environmental changes and represent important factors evaluating and diagnosing the skin conditions. This study investigates the critical roles of melanin and hemoglobin in skin-light interaction and combines spectral reflectance with single-point pigment values (collected by Mexameter MX18) to achieve the objective imaging skin color assessment. Feature wavelengths selected by the competitive adaptive reweighted sampling algorithm aligned well with narrow wavelength band designed by MX18, effectively removing redundant data while maintaining the model accuracy. Furthermore, seven machine learning methods were compared and evaluated, among which the stacked generalization model demonstrated the best performance (RMSEV = 14.23, $R_v^2=0.8634$ , RPD_v = 2.706 for melanin index; RMSEV = 31.74, $R_v^2=0.7505$ , RPD_v = 2.002 for erythema index). Finally, hyperspectral imaging technology enabled the visualization of skin pigment distribution, providing a rapid and non-invasive analytical tool for dermatological diagnosis and aesthetic evaluation.