Real-time activity classification in a wearable system prototype for knee health assessment via joint sounds.

Abstract

An algorithm for performing activity classification for a joint health assessment system using acoustical emissions from the knee is presented. The algorithm was refined based on linear acceleration data from the shank and the thigh sampled at 100 Hz/ch and collected from eight healthy subjects performing unloaded flexion-extension and sit-to-stand motions. The algorithm was implemented on a field-programmable gate array (FPGA)-based processor and has been validated in realtime on a subject performing two minutes of activities consisting of flexion-extension, sit-to-stand, and other motions while standing. When an activity is detected, the algorithm generates an enable signal for high throughput data acquisition of knee joint sounds using two airborne microphones (100 kHz/ch) and two single-axis gyroscope and accelerometer pairs (1 kHz/ch). This approach can facilitate energy-efficient recording of joint sound signatures in the context of flexion-extension and sit-to-stand activities from freely-moving subjects throughout the day, potentially providing a means of evaluating rehabilitation status, for example, following acute knee injury.