Neuromuscular Disease Detection Employing 1D-Local Binary Pattern of Electromyography Signals

Abstract

In this contribution, a novel technique for the detection of neuromuscular disorders is proposed employing a 1D-local binary pattern of electromyography signals. 1D-LBP is a local feature descriptor that is capable of identifying localized and sudden fluctuations present in EMG signals due to irregular firing patterns of motor neurons which is rooted in the physiology of the neuromuscular diseases. In the present contribution, initially, the 1D-LBP technique was applied on healthy, myopathy and amyotrophic lateral sclerosis EMG signals to obtain their respective LBP codes. The histogram of occurrence of LBP codes of different types of EMG signals was subsequently used as features to classify EMG signals using support vector machines (SVM) classifier. To reduce the size of the feature dimension, the performance of the proposed method was further evaluated using uniform 1D-LBP. Two binary classification problems were performed and investigations revealed that both conventional and uniform 1D-LBP returned very high detection accuracies for both problems, which can be potentially implemented for real-time neuromuscular disease detection.