Skin disorders represent a major global health burden and remain diagnostically challenging due to symptom heterogeneity, overlapping clinical features, and limitations of conventional approaches such as visual inspection and histopathology. Diffuse optical spectroscopy and imaging (DOSI) provides a noninvasive method for in vivo interrogation of skin physiology by quantifying light‐tissue interactions. When combined with machine learning (ML), these techniques enable automated, quantitative, and real‐time skin assessments, thereby improving diagnostic accuracy and minimizing observer bias. This review critically examines clinical studies on both conventional and ML‐integrated DOSI methods for dermatological applications, with a primary focus on the past decade and inclusion of earlier foundational work where appropriate. The literature is organized by disease category and imaging modality, with detailed evaluation of system designs, spectral biomarkers, preprocessing strategies, and ML models, including classical classifiers to deep and probabilistic architectures. Key challenges in standardization and clinical integration are discussed, alongside emerging opportunities in multimodal imaging, robust ML translation, and point‐of‐care deployment.
Hussain et al. (Fri,) studied this question.