Anterior cruciate ligament (ACL) injuries are common in athletes and often result in long-term complications such as altered gait and an increased risk of osteoarthritis. Identifying gait abnormalities after ACL reconstruction is crucial for optimizing rehabilitation. This study investigates the use of machine learning (ML) models to distinguish between athletes with ACL injuries and healthy individuals using spatiotemporal gait data collected during treadmill running. The Classifiers tested include the K-nearest neighbors (KNN), gradient boosting machine (GBM), XGBoost (XGB), random forest (RF), support vector machine (SVM), and multilayer perceptron (MLP). KNN demonstrated superior performance, achieving the highest balanced accuracy, precision, recall, and F1 score (0.98), with an area under the curve (AUC) of 0.99. Feature importance analysis revealed that temporal gait variables, particularly stride time, stance time, swing time, and anterior–posterior velocity, were the most critical predictors across all the models. These findings align with previous research showing post-ACL reconstruction gait changes. ML models, particularly KNN, show strong potential for detecting subtle gait differences, which have important clinical implications. Early detection of re-injury risk and personalized rehabilitation interventions could improve gait coordination and reduce the long-term risk of osteoarthritis associated with ACL injuries.
Tavakoli et al. (Thu,) studied this question.