IMU-augmented Patient-related Outcome Measure for Knee Arthroplasty Patients

Abstract

Purpose The aims of this study are (1) to derive a set of time series features to predict the total knee contact force during gait from two ankle-mounted inertial measurement units (IMUs); and (2) to explore the best combination of PROMs and IMU-derived, biomechanical measures for the prediction of postoperative quality of life (QoL) and reduce the inconsistency between biomechanical and PROMs. Methods Synced motion capture (optical and inertial) data were collected from four healthy participants to obtain a suitable time-series feature set that relates to knee contact forces (KCF). Then, using data from 28 patients during overground walking, we generated linear and random-forest regressor models for estimating QoL. These models were evaluated using temporal cross-validation. Results Overall, features from the IMUs could predict the total KCF (R 2 > 0.90, RMSE = 0.14, %BW). A machine learning model was trained on the data obtained during the recovery (up to 1 year) to predict EQ-5D-5 L score (patient QoL). The models were able to predict within a margin of 1.81–4.40 units of the actual score 95% of the time. Conclusion The presented study derives surrogate measures of total knee contact force (KCF) using ankle-mount IMUs. We identified the best combination of PROMs and IMU-derived measures for postoperative QoL at four time-points. The optimal metrics varied for each time point: OKS only, knee kinematics with OKS, KCF with OKS, and KCF only. Further data collection will be required to compile a comprehensive dataset to generate a robust predictive model.

Bibliographic Information