Predicting Return-to-Sport Timeline: Classifying Anterior Cruciate Ligament Health Levels Post-Reconstruction Surgery Using Convolutional Neural Networks

Sports-related anterior cruciate ligament (ACL) ruptures are linked to abnormal muscle activation patterns, extending treatment and delaying athletes’ return to sports (RTS) post-ACL reconstruction (ACLR). Given its prevalence among athletes, accurately estimating safe RTS time is crucial. This study introduces a deep convolutional neural network (DCNN) designed to classify ACL health levels in injured athletes, aiding in RTS estimation. Ten ACLR athletes and fourteen healthy controls participated. We defined three ACL health levels: healthy, six months post-ACLR, and nine months post-ACLR. Surface electromyography (sEMG) signals were recorded from five knee muscles during single-leg drop landing (SLDL) and single-leg forward hopping (SLFH) tasks. The time-frequency Pseudo-Wigner-Ville distribution (PWVD) calculated energy distribution patterns of non-stationary sEMG signals, generating heatmap images. The PWVD, preferred for its superior time-frequency localization over wavelet transform, processed these heatmaps, which were then analyzed by the custom-designed DCNN. The DCNN classifier achieved a 90% accuracy in categorizing ACL health levels. Additionally, it calculated the probability of data points belonging to specific classes, estimating safe RTS time for athletes. This study demonstrates the potential of DCNN classification using sEMG signals to estimate RTS time, providing valuable insights into ACLR recovery in athletes.