Autofluorescence and deep learning in early disease detection: biological foundations, clinical applications, and future directions.

Autofluorescence (AF) imaging enables label-free visualization of tissue metabolism and microenvironmental alterations, while deep learning (DL) provides powerful tools to decode its complex optical signatures. Their integration has emerged as a promising framework for functional and biologically informed disease assessment. Recent studies demonstrate that AF-DL approaches improve lesion detection, intraoperative guidance, and early therapeutic response evaluation across multiple organ systems. By leveraging multidimensional spectral, temporal, and spatial features, DL mitigates the intrinsic variability and limited specificity of standalone AF imaging. This review summarizes current diagnostic, prognostic, and surgical applications of AF-DL integration, with particular emphasis on model interpretability, generalizability, and biological relevance. Key challenges, including device dependence, dataset heterogeneity, annotation burden, and regulatory considerations, are critically discussed. Finally, future directions are proposed toward standardized acquisition, prospective multicenter validation, and clinically integrated workflows. By bridging intrinsic tissue biochemistry with data-driven intelligence, AF-DL integration offers a new class of functional imaging biomarkers with significant potential for precision diagnosis, surgery, and treatment monitoring.