Related Experiment Video
Updated: Jul 12, 2025
10:25
Brain Infarct Segmentation and Registration on MRI or CT for Lesion-symptom Mapping
Published on: September 25, 2019
48.1K
A Multi-Graph Cross-Attention-Based Region-Aware Feature Fusion Network Using Multi-Template for Brain Disorder
IEEE Transactions on Medical Imaging
|October 24, 2023
Summary
This study introduces a new multi-graph network for brain disorder diagnosis using resting-state fMRI. The novel approach improves accuracy by analyzing both static and dynamic brain connectivity across multiple templates.
Area of Science:
- Neuroscience
- Medical Imaging
- Artificial Intelligence
Background:
- Resting-state functional magnetic resonance imaging (rs-fMRI) is crucial for diagnosing brain disorders using functional connectivity (FC) networks.
- Existing methods often use a single template, limiting the capture of complex brain connectivities and neglecting complementary static/dynamic network information.
- Suboptimal diagnostic performance arises from ignoring functional divergence across brain regions.
Purpose of the Study:
- To propose a novel Multi-Graph Cross-Attention based Region-Aware Feature Fusion Network (MGCA-RAFFNet) for enhanced brain disorder diagnosis.
- To leverage multi-template analysis for a more comprehensive understanding of brain connectivity.
- To integrate static and dynamic brain network information and account for regional functional divergence.
Main Methods:
- Utilized multi-template brain parcellation to define Regions of Interest (ROIs).
- Developed a Multi-Graph Cross-Attention Network (MGCAN) with static and dynamic graph convolutions to analyze multi-template data.
- Implemented a Dual-View Cross-Attention (DVCA) mechanism for complementary information acquisition and a Region-Aware Feature Fusion Network (RAFFNet) for feature discrimination.
Main Results:
- The MGCA-RAFFNet demonstrated superior performance in diagnosing mild cognitive impairment (MCI) and autism spectrum disorder (ASD) on ADNI-2 and ABIDE-I datasets.
- The multi-template approach effectively captured complex brain interaction patterns.
- The proposed fusion network significantly improved feature discrimination by considering static-dynamic relationships within brain regions.
Conclusions:
- The MGCA-RAFFNet offers a significant advancement in brain disorder diagnosis by integrating multi-template, static-dynamic, and region-aware analyses.
- This novel network architecture effectively addresses the limitations of single-template methods and enhances diagnostic accuracy.
- The findings suggest a promising direction for developing more sensitive and reliable neuroimaging-based diagnostic tools.
