Explainable machine learning identifies ultrasound muscle thickness and strength as key predictors of lower limb function in older women

Maintaining lower limb function is essential for independence in older adults. In this study, we examined whether measurements of the thickness of the quadriceps via ultrasound combined with information on muscle mass and grip strength could be used to predict lower limb function in older women. Forty older women (mean age 74.6 ± 5.4 years) underwent clinical assessment including measurements of quadriceps thickness via ultrasound, adjusted skeletal muscle mass index, grip strength, cognitive testing, and the Short Physical Performance Battery (SPPB). Pearson correlations and multivariable linear regression were used to examine associations with the SPPB. Machine learning models (elastic net, random forest, and gradient boosting) were trained with 5-fold cross-validation. SHapley Additive exPlanations (SHAP) analysis was performed on the best-performing tree-based model. Quadriceps thickness (r=0.71), grip strength (r=0.47), and adjusted skeletal muscle mass (r=0.43) demonstrated strong positive associations with results of the SPPB, whereas age was inversely correlated (r=−0.43). Elastic net and random forest models achieved the highest predictive performance (R 2 ≈ 0.56–0.58) and outperformed linear regression. SHAP analysis consistently identified quadriceps thickness, grip strength, and muscle mass as dominant predictors of mobility, with minimal contribution from BMI. Explainable machine learning identified ultrasound-derived muscle thickness, grip strength, and skeletal muscle mass as key determinants of lower limb function in older women. These findings support the integration of portable ultrasound and simple strength testing with interpretable machine learning models to facilitate mobility risk screening and guide individualized rehabilitation strategies in geriatric care.