Change Point Detection in Knee Acoustic Emissions using the Teager Operator: A Preliminary Study in Patients with Juvenile Idiopathic Arthritis

Abstract

Acoustic emission events (clicks) produced by the knee during movement have a variety of causes but can be classified as physiologic or pathologic. In this paper, we present a pilot study investigating detection and classification of physiologic and pathologic clicks in knee acoustic emissions from 4 subjects with juvenile idiopathic arthritis (JIA) and 4 control subjects. First, the signals are filtered and spectral noise suppression is applied. Then, the clicks are located using the Teager energy operator and are extracted from the main signals. Several time and frequency domain features are extracted from each click. Using a random forest classifier, the clicks are categorized as “physiologic” or “pathologic”. In our dataset, we found an accuracy, sensitivity and precision of 94.3%, 93.3% and 96.6%, respectively, in correctly attributing these clicks to their respective classes. Similarly, the area under the receiver operating characteristics (ROC) curve is calculated as 0.98. The proposed click detection and classification pipeline may be used as an objective guide for knee health assessment in future work.