Evaluating pediatric peripheral neuromuscular disorders using deep neural networks on electrodiagnostic data

Area of Science:

• Medical Informatics
• Computational Neuroscience
• Neurology

Background:

• Neuromuscular disorders encompass a range of conditions affecting nerves and muscles.
• Accurate diagnosis is crucial for effective patient management.
• Electrodiagnostic tests provide valuable data for neuromuscular disorder assessment.

Purpose of the Study:

• To evaluate the efficacy of deep neural networks (DNNs) in predicting neuromuscular disorders.
• To compare DNN performance against traditional clinical assessment methods.
• To explore the potential of AI in enhancing diagnostic accuracy for neuromuscular conditions.

Main Methods:

• A dataset of electrodiagnostic and clinical information from intensive care patients over 10 years was utilized.
• Patients were classified into six diagnostic groups, including neuropathy, myopathy, and critical-illness neuromyopathy.
• A deep neural network was trained on the collected electrodiagnostic data.

Main Results:

• The trained neural network achieved promising validation results with 0.92 accuracy.
• High performance metrics were observed, including ROC-AUC of 0.99 and Precision Recall AUC of 0.97.
• Confusion matrix and positive predictive value analysis indicated high diagnostic performance, with diagonal values exceeding 0.82.

Conclusions:

• Deep neural networks demonstrate significant efficacy in predicting neuromuscular disorders based on electrodiagnostic data.
• The AI model's performance was found to be comparable to expert clinical assessment.
• Further development with larger datasets could establish DNNs as a reliable tool for neuromuscular diagnoses.