Work

Deep-learning–enabled spatial frequency domain imaging of the spatiotemporal dynamics of skin physiology

Year: 2025

Type: article

Abstract: Significance: . Aim: We aim to develop deep-learning-enabled spatial frequency domain imaging (SFDI-net) for real-time large FOV imaging of the optical, structural, and physiological properties and demonstrate its application for probing the spatiotemporal dynamics of skin physiology. Approach: SFDI-net, based on mapping of a two-layer structure into an equivalent homogeneous medium for spatially modulated light and with a convolutional neural network architecture, produces two-dimensional maps of optical, structural, and physiological parameters for bilayered tissue, including cutaneous hemoglobin concentration, oxygen saturation, scattering properties (reduced scattering coefficient and scattering power), melanin content, surface roughness, and epidermal thickness, with visible spatially modulated light at the camera frame rate. Results: Compared with traditional approaches, SFDI-net achieves a real-time inversion speed and significantly improves image quality by effectively suppressing noise while preserving tissue structure without oversmoothing. We demonstrate the application of the SFDI-net for monitoring the spatiotemporal dynamics of forearm skin physiology in reactive hyperemia and rhythmic respiration and reveal their intricate patterns in hemodynamics. Conclusions: Deep-learning-enabled spatial frequency domain imaging and SFDI-net may offer insights into the cardiorespiratory system and have promising clinical utility for disease diagnosis, surveillance, and therapeutic assessment. Future hardware and software advancements will bring SFDI-net to clinical practice.

Source: Journal of Biomedical Optics

DOI: 10.1117/1.jbo.30.4.046008