Classification of Muscle Inertial Motion and Electromyographic Activity Integration to Improve Accuracy in Pattern Recognition

Abstract

ABSTRACT Introduction Over the years, several studies have been published reporting the use of distinct sources of information used for pattern recognition that can be translated into commands to control human-machine interface system, for example, electromyography (EMG), pressure sensors, and accelerometers. Studies using muscle motion patterns and its combination with EMG in the context of pattern recognition for evaluation of the muscles and human-machine interface system in able-bodied individuals and limb-absent subjects are scarce. Material and Methods In this context, this research presents the assessment of the classification of patterns formed by features extracted from both muscle motion and electromyographic signals. Data sets were collected from both arms of five unilateral transradial limb-absent subjects and seven able-bodied subjects in the control group. The features from the EMG and the muscle motion such as amplitude, frequency, predictability, and variability of the signals were estimated. Results The results were presented in terms of the sensitivity, specificity, precision, and accuracy of the classifier. The combination of both measurements, EMG and muscle motion, defined the six basic movements for limb-absent subjects within an accuracy of 98% ± 1% for the sound forearm against 96% ± 4% for the amputated forearm. Conclusions For future work, it is expected that the strategy of classification and the combination of inertial and electromyographic activity will be used in actual scenarios for the controlling of artificial limbs and other applications related to human-machine interaction. Clinical Relevance The use of inertial sensors may increase the usability and accuracy of systems used for diagnosing, training, therapy, or controlling devices such as orthoses and prostheses.