Classification of Post-Anterior Cruciate Ligament Reconstruction Running Dynamics using Non-Traditional Features.

Despite extensive rehabilitation, nearly half of all post-anterior cruciate ligament reconstruction (ACLR) individuals are unable to perform dynamic tasks at the level they did prior to their injury. This inability can be attributed to unresolved neuromuscular deficits that manifest as altered limb dynamics. While traditional discrete metrics; such as peak vertical ground reaction force (vGRF) and peak knee flexion angle, have been used to successfully differentiate between healthy and pathological running dynamics, recent studies have shown that non-traditional metrics derived from autoregressive (AR) modeling and Smoothed Pseudo Wigner-Ville (SPWV) analysis techniques can also successfully delineate between healthy and pathological populations and could potentially possess greater sensitivity than the traditional metrics. Thus, the objective of this study was to compare the performance of classification models generated from traditional and nontraditional metrics collected from healthy controls and post-ACLR individuals during a running protocol. We hypothesized that the non-traditional metric-based classification model would outperform the traditional metric based model. Thirty-one controls and 31 post-ACLR individuals performed a running protocol from which the traditional metrics - peak vGRF, linear vGRF loading rate and peak knee flexion angle - and nontraditional metrics - dynamic vGRF ratio, AR model coefficients, and a SPWV derived low frequency-high frequency ratio - were extracted from vGRF and knee flexion running waveforms. The results indicated that a fine Gaussian SVM classification model derived from the non-traditional metrics had an accuracy of 87%, specificity of 83% and sensitivity of 61% and it outperformed the classification model derived from traditional metrics. These findings indicate that additional, valuable information can be ascertained from non-traditional metrics that evaluate waveform dynamics. Additionally, it suggests that this or similar models can be used to track the restoration of healthy running dynamics in post-ACLR individuals during rehabilitation.