Classification of Myopathy and Amyotrophic Lateral Sclerosis Electromyograms Using Bat Algorithm and Deep Neural...

Area of Science:

• Biomedical Engineering
• Signal Processing
• Neurology

Background:

• Electromyograms (EMG) record nerve and muscle activity, crucial for diagnosing neuromuscular disorders.
• EMG signals are non-stationary, posing challenges for accurate abnormality detection.
• Automated diagnostic tools require efficient feature extraction and classification of abnormal EMG signals.

Purpose of the Study:

• To develop and evaluate an efficient computer-aided system for diagnosing neuromuscular abnormalities using EMG signals.
• To compare the performance of deep neural networks (DNN) with conventional artificial neural networks (ANN) for EMG signal classification.
• To identify the optimal features and classification models for enhanced diagnostic accuracy.

Main Methods:

• Extracted time-domain and Wigner-Ville transformed time-frequency features from abnormal EMG signals.
• Utilized the bat algorithm for selecting the most relevant features.
• Developed and trained a deep neural network classifier on the selected feature subset.
• Compared the DNN performance against a conventional single-layer ANN.

Main Results:

• The deep neural network, using time-domain features, achieved 100% accuracy, sensitivity, and specificity in classifying EMG abnormalities.
• The conventional ANN model showed 83.3% accuracy, 100% sensitivity, and 71.42% specificity with time-domain features.
• The DNN demonstrated superior performance in classifying abnormal EMG signals compared to the ANN.

Conclusions:

• Deep neural networks, particularly with time-domain features, offer a highly effective approach for diagnosing neuromuscular system abnormalities.
• Feature selection using the bat algorithm enhances the efficiency of neural network classifiers for EMG signal analysis.
• The developed DNN model shows significant potential for improving automated diagnostic assistance tools for neuromuscular disorders.