Multi-Scale Dynamic Graph Learning for Brain Disorder Detection With Functional MRI.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Machine Learning

Background:

• Resting-state functional magnetic resonance imaging (rs-fMRI) is crucial for detecting brain disorders like autism spectrum disorder using machine/deep learning.
• Current methods often rely on functional connectivity networks (FCNs) but struggle to capture dynamic spatiotemporal changes and multi-scale topological information.

Purpose of the Study:

• To propose a multi-scale dynamic graph learning (MDGL) framework for automated brain disorder diagnosis.
• To address limitations in capturing spatiotemporal dynamics and single-scale network topology in rs-fMRI data.

Main Methods:

• The MDGL framework constructs multi-scale dynamic FCNs using multiple brain atlases.
• It employs multi-scale dynamic graph representation learning to capture spatiotemporal information from rs-fMRI data.
• Feature fusion and classification are performed across multiple scales.

Main Results:

• The MDGL framework effectively captures multi-scale spatiotemporal dynamic representations of rs-fMRI data.
• Experimental results on two datasets demonstrate superior performance compared to existing state-of-the-art methods.
• The framework successfully models multi-scale topological information for improved diagnostic accuracy.

Conclusions:

• The proposed MDGL framework offers a powerful new approach for analyzing rs-fMRI data in the context of brain disorder diagnosis.
• MDGL's ability to integrate multi-scale and dynamic information significantly advances the field.
• This method holds promise for more accurate and automated detection of neurological conditions.