[Research on migraine time-series features classification based on small-sample functional magnetic resonance imaging...

Area of Science:

• Neuroimaging analysis
• Computational neuroscience
• Medical diagnostics

Context:

• Migraine diagnosis relies on identifying subtle neuroimaging features.
• Traditional methods often use static image features, potentially overlooking temporal brain dynamics.
• Developing efficient diagnostic models for small-sample neuroimaging data is crucial.

Purpose:

• To develop and validate a novel method for migraine identification using brain time-series signals.
• To leverage temporal information in brain activity for improved diagnostic accuracy.
• To create a computationally efficient diagnostic model for migraine.

Summary:

• This study extracts regional average time-series signals from neuroimaging data of migraine patients and healthy controls.
• Group Independent Component Analysis and Dictionary Learning were used for brain segmentation and signal extraction.
• A bi-directional long-short term memory network modeled these time series, achieving 96.94% classification accuracy and a 0.98 AUC.
• The method effectively utilizes temporal information, reduces computational effort, and shows strong applicability for migraine diagnosis.

Impact:

• Provides a new, lightweight diagnostic model for migraine based on small-sample neuroimaging data.
• Enhances understanding of the neural discrimination mechanisms underlying migraine.
• Offers a promising approach for auxiliary diagnosis and personalized treatment strategies.