Deep Learning-Based Volition Detection and Action Potential Extraction for Fully Automated Diagnosis of Neuromuscular...

Area of Science:

• Neurology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Accurate diagnosis of neuromuscular diseases relies on interpreting needle electromyography (nEMG) signals.
• Manual analysis of nEMG signals for volition detection is labor-intensive and subjective.
• Automating this process can enhance diagnostic efficiency and accuracy.

Purpose of the Study:

• To develop a deep learning model for automated volition detection in nEMG signals.
• To improve the classification of neuromuscular diseases using automated nEMG analysis.
• To create a less labor-intensive diagnostic workflow.

Main Methods:

• Developed a deep learning model (nEMGNet) using 376 nEMG signals from 57 subjects.
• Validated the model on an independent dataset of 751 nEMG signals from 115 subjects.
• The model directly processed raw nEMG signals to extract volition signals automatically.

Main Results:

• The optimal segment length for volition detection was 0.060 s.
• Model-extracted volition signals improved classification performance compared to raw nEMG signals (AUROC: 0.806 for myopathy, 0.819 for neuropathy).
• Patient-wise AUROC improved by over 20% for all disease categories compared to physician-detected data.

Conclusions:

• The deep learning-based volition-detection model significantly enhances neuromuscular disease classification.
• The automated approach reduces diagnostic workload and improves efficiency.
• This technology offers a promising tool for objective and automated nEMG analysis.