Asynchronous Functional Brain Network Construction With Spatiotemporal Transformer for MCI Classification.

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Resting-state functional magnetic resonance imaging (rs-fMRI) is used to construct functional brain networks (FBNs) for disease diagnosis.
• Existing methods often assume synchronous brain activity and lack joint optimization for diagnosis, leading to variability and reduced accuracy.
• Modeling asynchronous functional connectivities (FCs) is crucial due to time lags in neural information flow.

Purpose of the Study:

• To develop a novel method for constructing and analyzing asynchronous FBNs using rs-fMRI.
• To address limitations of existing synchronous FC analysis and individual-level FBN construction.
• To improve diagnostic accuracy for functional brain diseases, specifically mild cognitive impairment (MCI).

Main Methods:

• A sliding-window-based method within a Transformer architecture is proposed to model spatiotemporal FCs.
• A novel approach learns common and individual FBNs adaptively, using common FBNs as prior knowledge to reduce variability.
• An integrated network enables joint construction and analysis of common and individual asynchronous FBNs for end-to-end training.

Main Results:

• The proposed method effectively models asynchronous functional connectivities.
• Adaptive learning of common and individual FBNs alleviates variability and focuses on disease-specific patterns.
• The integrated network demonstrates improved flexibility and discriminability in diagnostic tasks.

Conclusions:

• The developed method offers a more robust approach to FBN analysis by incorporating asynchronous FCs.
• The adaptive learning of common and individual FBNs enhances diagnostic accuracy for conditions like MCI.
• This integrated, end-to-end framework shows significant potential for clinical applications in brain disease diagnosis.