Answers are grounded in this paper and cite their sources.
This study aimed to characterize differences in facial and tongue CIELAB color distributions across five traditional Chinese medicine constitution categories and to assess changes in internal classification performance after adding facial features to tongue-based models. This single-center cross-sectional study included 779 university students aged 18–25 years. Standardized imaging provided 39 CIELAB features. Five constitution labels were derived from the 60-item Constitution in Chinese Medicine Questionnaire, which was not used as a model input. L2-regularized multinomial logistic regression and CatBoost were developed and evaluated using stratified nested cross-validation with five outer folds repeated three times and five inner folds. Macro-averaged F1 was the primary performance measure. Participant-level bootstrap resampling estimated 95% confidence intervals for performance, between-model differences, and modality increments. After multiplicity correction, 38 of 39 features differed significantly across categories. Macro-averaged F1 was 0.776 (0.739–0.808) for CatBoost and 0.752 (0.715–0.786) for logistic regression; the paired difference was 0.024 (− 0.005–0.053). Adding facial features to 21 combined tongue features yielded paired increments of 0.025 (0.002–0.048) and 0.030 (0.001–0.059), respectively. Tongue features provided most of the classification information, while facial features yielded modest internal performance gains. Combined features may inform constitution phenotyping and multimodal research; independent validation remains necessary.
No takes yet. Share an insight, caveat, or question.
Wei et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: