Machine learning approach for Migraine Aura Complexity Score prediction based on magnetic resonance imaging data

Area of Science:

• Neuroimaging
• Machine Learning
• Migraine Pathophysiology

Background:

• The Migraine Aura Complexity Score (MACS) system aids in studying migraine with aura (MwA) pathophysiology.
• Integrating advanced machine learning (ML) with detailed MwA profiling can yield novel insights.
• Structural cortical features hold potential for predicting MwA complexity.

Purpose of the Study:

• To evaluate ML algorithms for predicting MACS using structural cortical features.
• To enhance understanding of MwA pathophysiology through neuroimaging analysis.
• To identify specific cortical features associated with aura complexity.

Main Methods:

• Utilized a dataset of 340 MRI features from 40 MwA patients.
• Employed correlation tests and wrapper feature selection for ML model input.
• Applied Support Vector Machine (SVM), Linear Regression, and Radial Basis Function network for regression analysis.

Main Results:

• SVM achieved a high prediction accuracy (R-squared = 0.89) with wrapper feature selection.
• Identified specific cortical features, primarily in parietal and temporal lobes, linked to MwA complexity.
• Demonstrated significant changes in these cortical areas correlating with aura complexity.

Conclusions:

• SVM with wrapper feature selection proved most effective for predicting average MACS.
• The study provides a promising method for assessing MwA complexity.
• Findings lay groundwork for future MwA research, diagnosis, and treatment development.