A brain structure learning-guided multi-view graph representation learning for brain network analysis.

Area of Science:

• Neuroscience
• Computational Biology
• Machine Learning

Background:

• Resting-state brain networks reveal neural communication during rest.
• Analyzing these networks is crucial for understanding brain function but faces challenges like data heterogeneity and noise.
• Current methods struggle with accurately modeling complex brain connectivity.

Purpose of the Study:

• To develop an advanced brain network analysis method for improved mental disorder diagnosis.
• To address limitations in current approaches by integrating brain structure and multi-view graph learning.
• To enhance diagnostic accuracy (ACC) for mental health conditions.

Main Methods:

• Employed multi-view graph representation learning guided by brain structure.
• Utilized graph pooling to optimize network representations and reduce noise.
• Developed a multi-view graph convolutional network (GCN) with an attention-based adaptive module for view fusion.
• Constructed graph networks using the Smith atlas for superior resting-state network characterization.

Main Results:

• The proposed model achieved high performance on autism and cocaine use disorder datasets.
• Demonstrated superior accuracy compared to state-of-the-art methods.
• Achieved approximately 75% diagnostic accuracy (ACC) and 70% area under the receiver operating characteristic curve (AUC) on both datasets.

Conclusions:

• The combined approach of multi-view graph learning and brain structure learning effectively captures critical information in brain networks.
• This method enhances feature acquisition from diverse perspectives, leading to improved brain network analysis.
• The findings support the utility of this novel method for diagnosing mental disorders.