On Anterior Cruciate Ligament Injury Prevention: Utilizing Wearable Sensors to Track Potential High-Risk Knee Kinematics and Kinetics

Abstract

Participation in competitive sports is on the rise, and along with this comes an increase in sports-related injuries. While these injuries can happen to any part of the body, some of the most common are lower limb injuries, particularly to ligaments, with one of the most common being the anterior cruciate ligament (ACL). Within this work, we examined soft tissue overuse injury mechanisms using a novel tool for identifying potentially injurious loading cycles that occur in adolescent athletes (predominantly female athletes) with the long-term goal of prevention of this unintentional injury. We used a commercially available inertial measurement units (IMUs) to develop post-collection equations that could accurately and reliably estimate ground reaction force, knee moments, and knee angles; each a crucial factor in determining potential ACL injury. We partitioned this study into two distinct stages, each with specific purposes and goals. In the first stage, we used in vitro models to estimate our required metrics. Here, we used two IMUs and machine learning algorithms and compared the estimates to values measured by a motion capture and load cell system. We determined that the equations could estimate ground reaction force, knee moments, and knee angles with moderate accuracy and reliability. We additionally found that a tuned quaternion conversion method provided more accurate knee angle measurements than a commercially available sensor fusion algorithm. These findings allowed us to proceed to the second stage, human subject testing, where we developed modifications to the first-stage equations. Here we used four IMUs (two on each limb), the models developed in the first stage, and machine learning to construct a modified model; we then compared the estimates for these modified models to measurements of a motion capture system and force plate measurements. These modified models could estimate metrics with moderate accuracy; however, the knee moment reliability was not satisfactory. We then developed a confusion matrix to examine if these estimates could distinguish between low-risk and potentially damaging events, and results showed high classification accuracy, precision, specificity, and sensitivity. We also adjusted the quaternion conversion method for human subject use and found the tuned algorithm provided more accurate and reliable knee angle measurements. Our results show the potential of IMUs for tracking gameplay events and distinguishing between low-risk and potential-risk events. While we acknowledged several limitations within our work, we ultimately believe that IMUs may be able to aid coaches and sports physicians in preventing some athletes from sustaining ACL injuries.