Functional connectome fingerprinting: Identifying individuals and predicting cognitive functions via autoencoder.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Brain Imaging Analysis

Background:

• Functional network connectivity is crucial for understanding brain functions and serves as a unique identifier for individuals.
• Existing brain connectomes contain both shared and individual-specific information, but the latter's potential for predicting variability is underexplored.

Purpose of the Study:

• To enhance the uniqueness of individual brain connectomes by reducing shared neural activity.
• To investigate if refined connectomes can improve individual identification and predict cognitive behaviors.

Main Methods:

• Utilized an autoencoder network with sparse dictionary learning to process functional connectivity data.
• Applied the framework to Human Connectome Project (HCP) data for analysis.

Main Results:

• The refined connectomes achieved high accuracy (up to 99.5%) in distinguishing individuals.
• Improved prediction of high-level cognitive behaviors, including fluid intelligence, executive function, and language comprehension.
• Identified high-order association cortices as key regions for individual discrimination and behavioral prediction.

Conclusions:

• The proposed autoencoder framework effectively enhances individual uniqueness in brain connectomes.
• Refined functional connectivity networks offer a promising tool for studying cognition, behavior, and individual brain function.