A Machine-Learning Approach to Measure the Anterior Cruciate Ligament Injury Risk in Female Basketball Players

Abstract

Anterior cruciate ligament (ACL) injury represents one of the main disorders affecting players, especially in contact sports. Even though several approaches based on artificial intelligence have been developed to allow the quantification of ACL injury risk, their applicability in training sessions compared with the clinical scale is still an open question. We proposed a machine-learning approach to accomplish this purpose. Thirty-nine female basketball players were enrolled in the study. Leg stability, leg mobility and capability to absorb the load after jump were evaluated through inertial sensors and optoelectronic bars. The risk level of athletes was computed by the Landing Error Score System (LESS). A comparative analysis among nine classifiers was performed by assessing the accuracy, F1-score and goodness. Five out nine examined classifiers reached optimum performance, with the linear support vector machine achieving an accuracy and F1-score of 96 and 95%, respectively. The feature importance was computed, allowing us to promote the ellipse area, parameters related to the load absorption and the leg mobility as the most useful features for the prediction of anterior cruciate ligament injury risk. In addition, the ellipse area showed a strong correlation with the LESS score. The results open the possibility to use such a methodology for predicting ACL injury.