Feasibility of automatic knee kinematic feature learning for discriminating between individuals with and without a history of an anterior cruciate ligament reconstruction.

Abstract

Knee osteoarthritis is a degenerative joint disease that often develops following an anterior cruciate ligament (ACL) injury, even following surgical reconstruction (ACLr). This research evaluated whether biomechanical biomarkers, derived from wearable sensors, could differentiate people with an ACLr, who are at risk of early knee osteoarthritis, from healthy controls. Twelve participants with an ACLr and 19 controls participated. Continuous three-dimensional (3D) knee kinematics were captured using inertial measurement unit (IMU) sensors during sequential daily living tasks comprising sit-to-stand, walking, obstacle crossing, squatting, and stand-to-sit. Using a least absolute shrinkage and selection operator regression model, 468 knee time-series features were extracted to classify individuals with an ACLr from controls. Cohen's d effect sizes were calculated for features selected by the regression model to quantify between-group differences. The model achieved an accuracy of 80.7 %, with 92 % sensitivity and 74 % specificity. Seven features were retained from the model. The top two features with the greatest effect sizes when compared to controls were: a reduction in peak-to-peak knee axial rotation and maximum knee axial rotation angle (d = 1.35 and d = 1.31, respectively). The present study found that axial knee kinematics could serve as important biomarkers of an ACLr, potentially representing a modifiable feature for osteoarthritis treatment and prevention. These findings demonstrate the feasibility of early knee osteoarthritis detection using biomechanical biomarkers, providing preliminary evidence for the use of wearable sensors outside clinical settings and underscoring the possibilities for at-home monitoring.