Lower-limb malalignment—including genu varum and genu valgum—is a prevalent musculoskeletal problem that can accelerate joint degeneration and impair gait. We present a hybrid deep-learning framework that performs end-to-end detection, severity grading, and therapy triage from frontal radiographs or photographs. The pipeline couples YOLOv11x for rapid landmark localization of the hip, knee, and ankle with LegAlignNet for precise alignment analysis. Using vector geometry, the mechanical axis angle at the knee is computed from hip–knee and knee–ankle vectors to quantify deviation and assign severity (mild, moderate, severe). On a clinical test set of 100 images, the system achieved 89% accuracy, 91% precision, 88% recall, and 89% F1 for abnormality detection; severity grading reached a macro-F1 of 84% (per-class F1: mild 0.86, moderate 0.82, severe 0.83). For angle quantification, the model obtained a mean absolute error of 1.9° (95% CI 1.5°–2.3°) and RMSE of 2.6° relative to reference measurements. For mild–moderate cases, the framework generates evidence-based corrective exercise prescriptions, while severe cases are flagged for specialist referral. Together with the >90% training/validation accuracy observed during development, these results support a low-cost, scalable solution for early screening, remote assessment, and personalized rehabilitation planning.
Giahi et al. (Mon,) studied this question.