A dual track predictive model for assessing crow's feet aging across different clinical severity grades

Accurate assessment of crow's feet is limited by subjective scales and single-dimensional objective tools, impeding standardized evaluation. We recruited 350 healthy Chinese women and acquired macroscopic and microstructural parameters using VISIA-CR, DermaTOP®, and two-photon microscopy. Following expert clinical grading, we developed a dual-track modeling strategy that combines interpretable Ridge Regression and TPOT-optimized Support Vector Regression (SVR) to translate macroscopic parameters into clinical scores. SVR model interpretability was evaluated by SHapley Additive exPlanations (SHAP) analysis. Crow's feet severity increased non-linearly with age. Two-photon microscopy qualitatively supported the hypothesis that grade 4 corresponds to a phase of notable epidermal barrier disruption and grade 5 to pronounced dermal matrix disorganization. The Ridge Regression model achieved a robust overall fit (R² = 0.519). Macroscopically, the Stacked SVR model achieved superior predictive precision in the Early Stage (Grades 0-2) by capturing subtle non-linear textural shifts, whereas the Ridge Regression model maintained robustness from the Transition Stage (Grades 3-4) onwards. SHAP analysis supported the physiological plausibility of the non-linear model. This study clarifies the micro‑physiological basis of crow's feet aging and establishes a stage-specific dual-track predictive framework, providing an objective, standardized approach for skin aging assessment.