Myographic Information Enables Hand Function Classification in Automated Fugl-Meyer Assessment

Abstract

Automated systems for assessing upper extremity motor function post-stroke have been proposed as a higher resolution, more objective, or faster alternative to routine clinical assessment. These studies have been performed using either remote or wearable sensors attached to the subject. A common difficulty of these systems has been the quantification of hand function, a major component of post-stroke motor dysfunction. Mechanomyographic sensors have been untested in this field but have much potential as a practical way to quantify hand function, and as a complimentary modality to kinematic data for quantifying other arm movements. For this study a new automated system has been proposed which incorporates both kinematic and myographic information in the classification of arm motor function. Twenty-eight subjects with acute stroke were recruited and instructed to perform the motor function tasks of the FMA-UE. Motor features were pruned using the ReliefF feature selection algorithm and classification performed using a linear Support Vector Machine. Non hand function tasks classified using myographic data achieved a mean classification accuracy of 50.5% compared to 62.0% achieved using IMU data alone. For hand function tasks, a higher classification accuracy of 62.4% was achieved using myographic data alone.