Computing personalized brain functional networks from fMRI using self-supervised deep learning.

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Computational Neuroscience

Background:

• Functional MRI (fMRI) is crucial for understanding brain function.
• Characterizing individual brain functional networks (FNs) is complex.
• Current methods for computing FNs can be time-consuming and require external supervision.

Purpose of the Study:

• To develop a novel self-supervised deep learning (DL) method for computing personalized brain FNs directly from fMRI data.
• To assess the generalizability and predictive power of the DL-derived FNs.

Main Methods:

• A convolutional neural network with an encoder-decoder architecture was employed.
• The DL model was trained in a self-supervised manner to optimize functional homogeneity of FNs.
• The model was trained on Human Connectome Project fMRI data and tested on multiple datasets.

Main Results:

• The DL method successfully computed personalized FNs directly from fMRI data.
• The identified FNs demonstrated strong generalization across different datasets.
• Personalized FNs were predictive of individual differences in behavior, brain development, and schizophrenia status.

Conclusions:

• Self-supervised DL offers a rapid and generalizable approach for computing personalized brain FNs.
• This method enhances the characterization of brain functional neuroanatomy.
• Personalized FNs have significant potential for understanding individual variability and neurological conditions.