RESUMO
In post-ACLR individuals, gait variability often represents the presence of altered motor control. Quantifying variable limb loading is challenging, yet nonlinear analyses have been successful in detecting changes in gait variability due to altered motor control. Here, nonlinear metrics were derived and used to train multiple machine learning models to classify between healthy controls and post-ACLR individuals. The metrics were extracted from individuals' vertical ground reaction force data during a fast-walking trial as variable limb loading is exacerbated when the system is stressed and being challenged. It was hypothesized that effective differentiation between healthy control and post-ACLR individuals would be achieved using machine learning models derived from limb loading rate variability measures. Seventeen healthy control and fourteen post-ACLR participants with measured between-limb loading rate asymmetries completed the walking protocol. Ground reaction force data was collected on an instrumented treadmill where they performed walking trials at 1.5 m/s. Nonlinear limb loading rate measures extracted from the healthy controls and post-ACLR participants' data served as inputs to the models in order to train them to distinguish between the two states. A Decision Tree Classifier that utilized a bagging strategy was the best model for distinguishing between healthy control and post-ACLR participants. The model was successful in classifying participants, reporting an accuracy score of 73%, precision score of 100%, and an AUC score of 0.77, despite the smaller dataset. The ability to detect and classify post-ACLR loading rate variation has significant clinical implications, as these methods could be implemented in clinical settings to diagnose pathological limb loading dynamics and/or altered motor control.Clinical Relevance- This classification model can be easily integrated into the clinic to help diagnose pathological limb loading based solely on vertical ground reaction forces and can aid clinicians in providing data-driven metrics to help inform rehabilitation decisions.
