Toward dynamic tongue diagnosis: a conceptual framework for temporal phenotyping

Abstract

Digital tongue diagnosis has advanced considerably through image-based feature extraction and machine learning, yet its dominant analytical paradigm remains anchored in static, single-frame examination—an approach that inherently discards temporal information generated during the examination process, including stabilization dynamics, fluctuation patterns, and short-term physiological responsiveness. In this Perspective, we argue for a reorientation from static feature categorization towards temporal phenotyping. We propose a four-component analytical framework comprising: (i) standardized video acquisition with minimum technical specifications for clinical feasibility; (ii) temporal event alignment anchored to physiologically defined reference points such as protrusion onset and maximal extension; (iii) steady-state window identification using CIELAB ΔE-based criteria to isolate analytically meaningful intervals; and (iv) dynamic feature representation that distinguishes quantitative temporal metrics from clinically interpretable phenotypes. We further delineate the evidentiary requirements—biological plausibility, measurement reproducibility, analytical validity, and clinical utility—necessary to substantiate this paradigm shift, and outline a staged validation pathway from controlled laboratory studies to prospective clinical evaluation. Potential applications include longitudinal health monitoring, individualised assessment, and telemedicine integration. Despite substantial translational barriers, we contend that a temporally informed framework offers a more comprehensive and clinically meaningful foundation for next-generation digital tongue diagnosis systems.