Constructing brain functional network by Adversarial Temporal-Spatial Aligned Transformer for early AD analysis.

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Brain functional networks reveal neural activity and are crucial for analyzing early Alzheimer's disease (AD).
• Current methods for constructing functional connectivity networks from fMRI data rely on software toolboxes with subjective settings, potentially leading to errors and reduced performance in disease analysis.

Purpose of the Study:

• To propose a novel Adversarial Temporal-Spatial Aligned Transformer (ATAT) model for automated mapping of 4D fMRI data into functional connectivity networks.
• To enhance the accuracy and performance of early AD analysis and progression monitoring.

Main Methods:

• The ATAT model incorporates region-guided feature learning to focus on local brain region features.
• A spatial-temporal aligned transformer network adaptively adjusts features to capture global functional connectivity patterns.
• A multi-channel temporal discriminator is used to differentiate generated and real multi-region time series distributions.

Main Results:

• Experimental results on the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset demonstrated the effectiveness and superior performance of the ATAT model.
• The model showed strong capabilities in early AD prediction and progression analysis.

Conclusions:

• The ATAT model offers a novel approach to constructing functional connectivity networks, improving neuroscience exploration and clinical disease analysis.
• The model's ability to evaluate disease-related changes at different stages provides valuable insights into AD pathology.