Title : Advancing neuromuscular disorder diagnosis and adaptive electroceutical therapy through real-time electromyography and hybrid deep learning
Abstract:
Neuromuscular disorders are a class of conditions that deteriorate the nerves controlling voluntary skeletal muscle, contributing to progressive motor dysfunction, losses in mobility and coordination, and muscle atrophy. According to the National Institutes of Health, it is estimated that neuromuscular disorders collectively affect every 1 in 3,500 people. Current clinical diagnostic approaches involve nerve conduction studies and needle electromyography, which are invasive and have limited sensitivity in detecting early neuromuscular changes. Similarly, clinical therapeutic approaches, including physiotherapy, primarily manage symptoms and lack the ability to adapt to individual patient-specific physiology. This research presents a software framework that improves the management of neuromuscular disorders through biophysical modeling and biomedical signal processing algorithms. Surface electromyography signals are preprocessed to capture high-dimensional spatiotemporal features representative of time-series neuromuscular activity. A dual-stage artificial intelligence pipeline integrating a support vector machine (SVM) and a hybrid convolutional neural network—long short-term memory network (CNN-LSTM) was developed. The SVM is leveraged for low-latency muscle activation state detection, while the hybrid CNN-LSTM is employed for classifying spatiotemporal patterns associated with neuromuscular disorders. In parallel, a computational implementation of the Hodgkin-Huxley conductance model was engineered to simulate motor neuron and muscle dynamics under varying electrical stimulation parameters. The proposed framework was evaluated using a standard set of machine learning performance metrics. The SVM achieved a 94.4% classification accuracy with a 20 ms inference latency, while the hybrid CNN-LSTM achieved a 96.1% diagnostic accuracy (ROC-AUC = 0.97), demonstrating the implications of deep learning-driven computational neuroengineering to facilitate neuromuscular disorder diagnosis and adaptive functional electrical stimulation.
