Myo-Guide: A Machine Learning-Based Web Application for Neuromuscular Disease Diagnosis With MRI.

Area of Science:

• Biomedical Imaging
• Artificial Intelligence
• Neurology

Background:

• Neuromuscular diseases (NMDs) cause progressive muscle weakness and disability.
• Early diagnosis is crucial for NMD management and genetic counseling.
• Muscle MRI is valuable but interpretation is complex, hindering clinical utility.

Purpose of the Study:

• To develop a multi-study harmonization pipeline for muscle MRI data.
• To create an AI-driven diagnostic tool for identifying NMD-specific muscle involvement patterns.
• To improve the accuracy and accessibility of NMD diagnosis.

Main Methods:

• Standardized MRI fat content using a preprocessing pipeline.
• Trained ensemble XGBoost models on intramuscular fat, age, and sex.
• Utilized SHAP for pattern analysis and class-balanced metrics for evaluation.

Main Results:

• Harmonized 2961 MRI samples from 20 NMD types.
• Achieved 64.8% balanced accuracy and 90.2% top-5 accuracy.
• AI model surpassed expert clinicians in diagnostic ranking (75.0% top-3 accuracy).

Conclusions:

• AI in muscle MRI for NMD diagnosis is promising despite data scarcity.
• The developed framework enables advanced computational techniques for NMD research.
• The Myo-Guide platform offers global access to an AI tool outperforming expert diagnosis.