Brain fingerprinting and cognitive behavior predicting using functional connectome of high inter-subject variability

Area of Science:

• Neuroscience
• Brain Imaging
• Machine Learning

Background:

• Brain functional connectivity (FC) graphs serve as unique individual "fingerprints."
• Existing methods struggle to separate shared inter-subject information from individual-specific information in FC graphs.
• This limitation leads to suboptimal individual identification accuracy and cognitive behavior prediction.

Purpose of the Study:

• To develop a novel method for enhancing inter-subject variability in brain FC graphs.
• To improve individual identification accuracy from fMRI data.
• To enhance the prediction of cognitive behaviors based on refined FC graphs.

Main Methods:

• Proposed a method combining a conditional variational autoencoder (CVAE) network and a sparse dictionary learning (SDL) module.
• Embedded fMRI state information into CVAE's encoding and decoding processes to capture common features and enhance inter-subject variability.
• Utilized Human Connectome Project (HCP) data for experimental validation.

Main Results:

• Achieved high individual identification accuracies: 99.7% (rest1-rest2) and 99.6% (rest2-rest1).
• Task-task identification accuracies ranged from 94.2% to 98.8%.
• Identified Frontoparietal and Default networks as crucial for individual identification and showed improved cognitive behavior prediction.

Conclusions:

• The proposed CVAE with SDL framework effectively enhances inter-subject variability in FC graphs.
• The refined connectomes improve individual identification accuracy and cognitive behavior prediction.
• This approach offers a promising avenue for studying brain function, cognition, and behavior.