Multimodal MRI-based classification of migraine: using deep learning convolutional neural network.

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Migraine is a common neurological disorder often misdiagnosed using current criteria.
• Deep learning (DL) shows promise in medical image analysis for disease detection and classification.
• Automated methods are needed to improve migraine diagnosis accuracy.

Purpose of the Study:

• To evaluate DL for classifying migraine patients and controls.
• To differentiate between migraine with aura and migraine without aura using DL.
• To assess the efficacy of DL with resting-state functional magnetic resonance imaging (rs-fMRI) data.

Main Methods:

• Utilized DL methods with three rs-fMRI functional measures: amplitude of low-frequency fluctuations (ALFF), regional homogeneity (ReHo), and regional functional correlation strength (RFCS).
• Classified 21 migraine patients without aura, 15 migraineurs with aura, and 28 healthy controls.
• Compared an Inception module-based convolutional neural network (CNN) with a traditional support vector machine (SVM) and an AlexNet-based CNN.

Main Results:

• The Inception module-based CNN achieved over 86.18% accuracy, significantly outperforming the SVM (83.67%).
• The Inception module-based CNN demonstrated superior performance with 99.25% accuracy compared to AlexNet-based CNN.
• Regional functional correlation strength (RFCS) emerged as the optimal input measure among ALFF, ReHo, and RFCS.

Conclusions:

• Combining rs-fMRI functional measures with DL classification effectively distinguishes migraineurs from healthy individuals.
• DL frameworks offer a powerful approach for developing advanced systems to support clinical decision-making in neurology.
• This study highlights the potential of AI in improving the diagnosis of complex neurological disorders like migraine.