Hamstrings–quadriceps (H–Q) balance is central to knee stability during running, yet static indices often fail to capture task-specific dynamics. We propose a digital assessment framework that combines simulation-informed biomechanical features with machine learning in an in-silico proof-of-concept. Synthetic running data were generated for 160 virtual subjects across three speeds, yielding 573 trials. Features included dynamic H:Q ratio, knee-moment and stance-time asymmetry, vertical GRF asymmetry, co-contraction, timing, and variability. A Gradient Boosting classifier with isotonic calibration was trained using subject-wise cross-validation. Performance was evaluated with ROC-AUC, PR-AUC, balanced accuracy, F1 score, and Brier score; bootstrap resampling provided confidence intervals. Interpretability was examined through permutation importance, partial dependence plots, and error analysis. The framework showed robust classification, with ROC-AUC = 0.933 (95% CI 0.908–0.958), balanced accuracy = 0.943 (95% CI 0.924–0.962), PR-AUC = 0.918, F1 = 0.940, and Brier score = 0.056. Dynamic H:Q and knee-moment asymmetry were the top predictors, while partial dependence revealed biomechanically plausible U-shaped and monotonic effects. Digital assessment of H–Q imbalance using simulation-informed features and machine learning enables robust detection with transparent interpretation. Although based entirely on synthetic data, this framework provides a reproducible methodological baseline for future validation in sports medicine.
Mănescu et al. (Wed,) studied this question.