Brave Virtual Worlds: Implementation of Motion Capture Methodologies for Predictive and Preventative ACL Rehabilitation
Parikh, Dhyey, School of Engineering and Applied Science, University of Virginia
Kodama, William, Engineering Undergraduate, University of Virginia
The ability to track and monitor injurious biomechanical movement patterns has become a critical area of development in the fields of sports and rehabilitation. Lower extremity injurious movements have been linked as risk factors associated with anterior cruciate ligament (ACL) injuries. With an annual incidence rate of 200,000 to 400,000 cases in the US, ACL rupture is a common problem in younger populations participating in medium to high intensive activities. While surgical reconstruction is commonly done for such injuries, the ability restore normal joint function and mitigate long-term development of early onset osteoarthritis is difficult to understand. However, through the study of ACL injury mechanisms, there may be the development of a process to design training programs aimed at prevention. This involves objective observation of contributing motions that occur causing an ACL rupture. One common example is the dynamic knee valgus collapse, which is usually described as excessive medial collapse of the knee. The aim of the following project was to use a custom on-body motion-capture wearable to gather data on the dynamic knee valgus collapse. The data was then analyzed using various machine learning algorithms to determine the accuracy of classifying a knee valgus collapse versus a normal knee abduction including K-Nearest Neighbors, Logistic Regression, and a Recurrent Neural Network called the Long Short-Term Memory algorithm. The testing accuracy of each model was compared against that of the current research standard of detecting human movement patterns, the K-Nearest Neighbors model. The Long Short-Term Memory algorithm demonstrated the highest level of accuracy statistically significant at a p-value of 0.01891 with a Logistic Regression model demonstrating a mean time of classification of 0.0004 s. Thus, the use of an on-body motion capture wearable paired with machine learning based classification can provide an effective and cost-efficient methodology for ACL injury preventative care.
BS (Bachelor of Science)
ACL, motion capture, machine learning, preventative care, dynamic knee valgus collapse
All rights reserved (no additional license for public reuse)