Background/Purpose: Osteoarthritis (OA) of the knee is a leading cause of disability, healthcare expenditures, and is associated with diminished quality of life and increased mortality. Varus thrust is a pathologic gait pattern that is more prevalent in individuals with obesity and associated with incident and progressive knee osteoarthritis. Knee acoustic emissions (KAE) is a practical, point-of-care assessment of knee health that can inform several pathological states, including early and established knee OA. Whether KAE is sensitive to knee OA risk or prognostic factors, such as varus thrust, is unknown. This study aims to evaluate the accuracy of a KAE-based model for classification of knees with varus thrust.

Methods: We recruited 20 adult participants with obesity with or without frequent and persistent knee pain from the Minneapolis Veterans Affairs Medical Center (MVAMC), whose institutional review board approved the study. During motion analysis, KAE was measured from bilateral knees using four contact microphones placed medially and laterally on the superior and inferior aspect of the patella. The presence of varus thrust was assessed by two independent reviewers using videos of ambulation. Participants ambulated at self-selected slow, medium, and fast speeds. KAE from each condition were isolated from one another to specific phases of gait using 30 ms windows with 20 ms overlap. 73 spectral, temporal, and amplitude-based features were extracted per window to train an XGBoost classifier for varus thrust detection. Knees with a percentage of positively classified windows above an empirically derived optimal threshold were classified as thrust-positive. Performance was assessed via 5-fold cross validation to obtain average values for accuracy, sensitivity, and specificity.

Results: Twenty participants (13 males) were recruited, with a mean age of 54.5 years and average body mass index (BMI) of 39.2. Eleven participants reported frequent and persistent knee pain. Twenty-three knees were graded Kellgren–Lawrence ≥2. Aggregate Knee Injury and Osteoarthritis Outcome Score (KOOS) subscale scores were: Activities of Daily Living – 80, Sports– 62, and Quality of Life – 58. Twenty-five percent of knees had varus thrust. The KAE model had a sensitivity, specificity, and accuracy of 70%, 76%, and 82%, respectively.

Conclusion: Models trained with KAE can accurately classify knees with varus thrust. KAE likely contain information about the larger biomechanical status of the knee and future studies should assess whether acoustics can predict other pathologic biomechanics of the knee.

Figure 1. Workflow for measurement and extraction of KAE. A. Contact microphones and associated hardware for KAE measurement. B. Microphones and hardware mounted on participant’s knee. C. Assessment for varus thrust at mid-stance phase of gait during ambulation. D. Extraction of KAE features by biomechanically-defined gait cycle into feature matrix for training classification model. KAE: knee acoustic emissions.

Figure 2. Receiver Operating Characteristic Curve of KAE Classification Model for Varus Thrust. Model performance area under the curve was 0.75. ROC: Receiver operating characteristic; AUC: area under the curve.

Table 1. KAE Model Results

« Back to ACR Convergence 2025

ACR Meeting Abstracts - https://acrabstracts.org/abstract/evaluating-the-accuracy-of-knee-acoustic-emissions-classification-of-varus-thrust/