Interpretable Artificial Intelligence Analysis of Functional Magnetic Resonance Imaging for Migraine Classification:...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Deep learning models show promise for diagnosing neuropsychiatric disorders like migraine but lack interpretability, hindering clinical use.
• The
• black-box
• nature of these models impedes biomarker discovery and personalized treatment strategies.

Purpose of the Study:

• To evaluate explainable artificial intelligence (XAI) techniques combined with functional magnetic resonance imaging (fMRI) indicators for migraine classification.
• To identify optimal pairings of AI models and fMRI indicators for improved diagnostic accuracy.
• To assess XAI's potential in clinical settings by pinpointing discriminative brain regions associated with migraine.

Main Methods:

• Analysis of resting-state fMRI data from 64 participants (migraine patients and healthy controls).
• Extraction and classification of three fMRI metrics: amplitude of low-frequency fluctuation, regional homogeneity, and regional functional connectivity strength (RFCS).
• Utilized deep learning models (GoogleNet, ResNet18, Vision Transformer) and conventional machine learning methods (SVM, Random Forest) for classification, with XAI generating activation heat maps.

Main Results:

• The GoogleNet model with RFCS indicators achieved the highest classification accuracy (>98.44%) and an AUC of 0.99.
• RFCS indicators improved classification accuracy by approximately 8% compared to amplitude of low-frequency fluctuation.
• XAI-generated heat maps highlighted the precuneus and cuneus as the most discriminative brain regions for migraine.

Conclusions:

• XAI combined with fMRI brain region features offers visual explanations for migraine progression.
• Understanding AI decision-making processes through XAI has significant potential for improving clinical migraine diagnosis.
• This approach shows promise for enhancing diagnostic accuracy and developing novel diagnostic techniques for migraines.