Knee Osteoarthritis Classification System Examination on Wearable Daily-Use IMU Layout

Abstract

The diagnostic approach for knee osteoarthritis that draws on kinematic characteristics provides a solution other than imaging medicine. However, the gait-based kinematic analysis still requires a motion capture suit as a prerequisite to ensure a reliable calculation, which limits the daily screening of the end user. To further reduce the cost, in this paper we investigated a wearable inertial measurement unit (IMU)-based knee osteoarthritis classification system based on daily-use wearing IMU layout and machine learning approaches. The acceleration and angular velocity signal output from the IMU were used as the input data; the different features from the time and frequency domains were examined with different handcrafted feature classifiers, as well as the deep learning method. From the results, using three IMUs could reach a 0.82 area under the curve value, with a sensitivity of 86% and a specificity of 78%. The results showed that using daily IMU devices to establish a diagnostic system with an on-body sensor layout is feasible.