Associate Professor University of Michigan-Dearborn Dearborn, Michigan, United States
Introduction:: It has been estimated that each year approximately 1 ACL tear occurs in every 50-70 female athletes. Submaximal repetitive loading is known to cause cumulative micro-damage in soft tissues, most notably in the elbow. Recently, laboratory research has indicated that some noncontact ACL injuries that occur may be overuse injuries due to repeated strain on the ACL when landing at 3x-4x body weight. The purpose of this study was to determine whether inertial measurement units (IMU) could be used to identify loading cycles that may be injurious to the ACL if repeated excessively.
Materials and Methods:: Prior to commencement of the study, IRB approval was acquired. Nine volunteers were recruited and fitted with four IMUs and 46 reflective spherical markers and were asked to perform several actions that occur during multidirectional sports. Ground reaction force was measured using a force plate and knee moments were determined using inverse kinematics. The ‘Regression Learning’ app developed by MATLAB was used to apply several types of machine learning algorithms to estimate the ground reaction force and the triaxial moments. Features used in the algorithms included height, mass, thigh circumference, distances to the femoral and tibial IMU, the linear accelerations and angular velocities measured by the IMUs and estimated values obtained through previously developed algorithms using cadaveric testing. Because the purpose was to distinguish between low- and high-risk events, a confusion matrix was constructed to gauge the sensitivity and specificity to correctly identify a class. Class values were based on a combination of ground reaction force (Low Risk: < 3x BW; High Risk: >3x BW), abduction moments (Low Risk: < 150 Nm; High Risk >150 Nm), and rotation moments (Low Risk: < 45.5 Nm; High Risk: >45.5 Nm).
Results, Conclusions, and Discussions:: Events that were considered potentially injurious loading cycles (possible risk events) were estimated with 78.6% sensitivity and 98.1% specificity (Figure 1). The positive predictive value was 78.6% and the negative predictive value was 98.1%. Previous studies have used basic regression modeling techniques to establish correlations between data obtained from an IMU and metrics measured by a motion capture system. However, our focus was on the development of models that could estimate those metrics and classify them as either high or low risk to potential injury. IMUs have the potential to be used to estimate ground reaction force and knee moments in order to classify events as potentially injurious. This could be important for on-field injury tracking.
Acknowledgements (Optional): : The authors would like to thank the research participants as well as Regina Arriola and Ruchika Tadakala for assisting with data collection.
