Select then fusion: An effective multi-atlas brain network analysis method with sparse and uncertain mechanism

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Multi-atlas brain networks provide a more detailed understanding of brain disorders than single atlases.
• Traditional methods struggle with information redundancy and uncertainty inherent in multi-atlas approaches.

Purpose of the Study:

• To develop an advanced multi-atlas brain network analysis method for improved brain disorder identification.
• To address limitations of existing fusion techniques in handling complex multi-atlas data.

Main Methods:

• Constructed multi-atlas brain networks from functional magnetic resonance imaging (fMRI) data using diverse atlases.
• Employed an attention-enhanced module for feature learning and a brain region selection module to identify disease-relevant areas.
• Implemented an uncertain fusion module to manage atlas uncertainty and achieve evidence-level results.

Main Results:

• The proposed Sparse and Uncertain Fusion Neural Network (SUFNN) demonstrated superior performance in identifying brain disorders.
• SUFNN effectively managed information redundancy and uncertainty, outperforming state-of-the-art methods on the SRPBS dataset.

Conclusions:

• The SUFNN method offers a robust and effective approach for multi-atlas brain network analysis.
• This technique holds significant potential for advancing the diagnosis and understanding of brain disorders using neuroimaging data.