On Neuromuscular Disease Detection from Electromyograms using Convolutional and Recurrent Neural Networks

This work presents a comparative study of six architectures for the detection of neuromuscular diseases from electromyograms (EMG) using deep learning techniques. By leveraging convolutional, recurrent, and hybrid convolutional-recurrent neural networks, we aim to improve the accuracy and reliability of classifying normal, myopathic, and neuropathic health conditions. Considering the EMGLAB database, which includes EMG recordings from both healthy individuals and patients with myopathy and amyotrophic lateral sclerosis, we propose a series of models that are rigorously evaluated through bi-class and multi-class classification tasks. The experimental results show that the proposed models, particularly CONVLSTM, a hybrid convolutional-recurrent model, achieves superior performance with classification accuracy of $99 \%$ in distinguishing among different neuromuscular conditions. This work offers significant insights into the application of deep learning in the accurate classification of complex EMG signals for neuromuscular diagnostics.