AI-Empowered Dual-Wavelength NIR System for Skin Hydration Detection

Skin hydration is a contributing factor to skin health and serves as an indirect measure of skin barrier function. Measuring skin hydration is important for dermatology and aesthetic research. This paper proposes a two-wavelength optical approach for detecting skin hydration in seventeen healthy adult volunteers. The developed system used market-available near-infrared (NIR) light-emitting diodes (LEDs) with center wavelengths of 1200 nm and 1450 nm, and an NIR sensor to detect reflected light. Experiments were conducted on the anterior and posterior forearms of the volunteers before and after the topical application of a moisturizing cream. A 1D convolutional neural network (CNN) was modified and optimized for skin hydration classification. The learning capacity of this model was improved by fine-tuning the network filter numbers and four training hyperparameters, namely the learner type, epoch number, minibatch size, and initial learning rate, using a grid search. This base model was also ensembled with four traditional machine learning (ML) classifiers, Support Vector Machine (SVM), K-Nearest Neighbour (KNN), Decision Tree (DT), and Naïve Bayes (NB), whose performances were compared. The results from the nested cross-validation experiments revealed nonmonotonic but comparable classification performance of the 1D CNN across varying k-fold values, with mean classification accuracy ranging from 79% to 82%. The integration of this model with the probabilistic NB model was shown to outperform other ensemble classifiers in all performance measures used, with the highest accuracy, precision, sensitivity, specificity, and F1 scores given by 89.23 ± 8.55%, 91.11 ± 6.97%, 89.51 ± 8.76%, 94.58 ± 4.45%, and 89.18 ± 8.50%, respectively. These preliminary findings support the feasibility of the proposed lightweight system as an alternative for non-invasive, non-contact skin hydration analysis. The prediction performance can be further improved by incorporating longer NIR wavelengths to capture more water-related spectral information for analysis.