Deep neural network assisted diagnosis of time-frequency transformed electromyograms

• Published in: Multimedia tools and applications Vol. 79; no. 15-16; pp. 11051 - 11067
• Main Authors: Bakiya, A., Kamalanand, K., Rajinikanth, V., Nayak, Ramesh Sunder, Kadry, Seifedine
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.04.2020; Springer Nature B.V
• Subjects: Artificial neural networks; Classification; Classifiers; Computer Communication Networks; Computer Science; Data Structures and Information Theory; Diagnosis; Electromyography; Fourier transforms; Frequency domain analysis; Multimedia Information Systems; Muscles; Neural networks; Particle swarm optimization; Special Purpose and Application-Based Systems; Electromyograms; Feature selection; Deep neural networks; Shallow neural networks; Transformation techniques; Time-frequency features
• ISSN: 1380-7501; 1573-7721
• DOI: 10.1007/s11042-018-6561-9

Electromyograms (EMG) are recorded electrical signals generated from the muscles and these signals are closely interrelated with the muscle activity and hence are useful for the investigation of neuro-muscular disorders. The feature mining, feature collection and development of classification systems are greatly significant steps in the differentiation of normal and abnormal EMG signals to evaluate the abnormality. In this work, time-frequency domain based features of regular, myopathy and Amyotrophic Lateral Sclerosis (ALS) EMG signals were extracted from four different techniques namely Stockwell-Transform (ST), Wigner-Ville Transform (WVT), Synchro-Extracting Transform (SET) and Short-Time Fourier Transform (STFT). The Particle Swarm Optimization (PSO) with fractional velocity update technique was implemented for feature reduction. Further, the classifier based on the Deep Neural Networks (DNN) was developed by employing the features selected using fractional PSO. Finally, the performance of the DNN was compared with that of the Shallow Neural Network (SNN) classifier. Results of this work demonstrate that, the performance measure of the DNN classifiers is higher than that of the SNN classifier. This work appears to be of good clinical significance since efficient classification techniques are required for the development of robust neuro-muscular diagnosis systems.