Automatic Detection of Compensatory Movement Patterns by a Pressure Distribution Mattress Using Machine Learning Methods: A Pilot Study

Abstract

For stroke patients with hemiplegia, reaching with the paretic arm is often associated with compensatory movements due to limited active arm movement and a loss of interjoint coordination. Detecting common compensatory movement patterns, such as excessive trunk displacement and scapular elevation, is critical for improving the motor function of the paretic arm. Existing compensatory movement pattern detection methods, including sensor-based systems and camera-based systems, suffer from object obstruction and require complex setups. In this paper, a compensatory movement pattern detection system using a pressure distribution mattress is presented. This method is a novel approach to detect compensations and has observed advantages; it is simple, unobtrusive and low cost. Fifteen healthy participants with no motor impairments performed three reaching tasks (back-and-forth, side-to-side, and up-and-down reaching) in a normal pattern and in compensatory movement patterns (trunk rotation and lean-forward, and scapular elevation). Pressure distribution data of all motions were recorded and processed to generate a group of features (average sensor values, the lateral center of pressure, longitudinal center of pressure, the ratio of left-side to right-side pressure, and the ratio of front-side to back-side pressure) reflecting the information of each predefined pattern. Four machine learning methods were implemented to detect compensatory movement patterns and showed good reliability and precision. Both k-nearest neighbor (kNN) and support vector machine (SVM) classifiers have achieved an excellent classification performance (F1-score = 0.934) in detecting compensation during all reaching tasks. For the multiclass classification of compensatory movement patterns, the SVM classifier exhibited a good classification performance for trunk lean-forward (F1-score = 0.933), scapular elevation (F1-score = 0.881), and trunk rotation (F1-score = 0.854) and outperformed previous reports. The study results provide initial evidence of a pressure distribution mattress for detecting compensatory movement patterns. Future work needs to test the approach on stroke survivors to verified the feasibility and validity.