Deep Learning outperforms physicians in myopathy and neuropathy classification based on Needle Electromyography...

Area of Science:

• Neurology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Needle electromyography (nEMG) is crucial for diagnosing neuromuscular disorders.
• Current nEMG analysis is labor-intensive and susceptible to human bias, potentially leading to diagnostic inaccuracies.

Purpose of the Study:

• To validate a deep learning (DL) system for classifying nEMG signals into normal, myopathy, and neuropathy categories.
• To compare the diagnostic performance of the DL system against experienced electromyographers.

Main Methods:

• A DL model was trained and validated on 376 nEMG signals from 57 patients using nested k-fold cross-validation.
• The DL model utilized minimal preprocessing for classifying patients.

Main Results:

• The DL model achieved higher median classification performance (accuracy 0.70, precision 0.70, sensitivity 0.70, specificity 0.85) compared to six electromyographers (accuracy 0.55, precision 0.60, sensitivity 0.54, specificity 0.78).
• Model interpretability confirmed classification based on relevant signal features.
• Despite higher overall accuracy, the DL model had more unanimously misclassified cases than physicians.

Conclusions:

• Deep learning presents a fast, accurate, and practical approach to assist physicians in diagnosing neuromuscular diseases via nEMG analysis.
• The validated DL system shows potential to improve diagnostic efficiency and accuracy in clinical practice.