Discovering individual fingerprints in resting-state functional connectivity using deep neural networks

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Interindividual differences in resting-state functional MRI (rfMRI) analysis present challenges.
• Identifying unique brain patterns is crucial for understanding brain function.

Purpose of the Study:

• To employ deep neural networks (DNNs) for individual identification using rfMRI.
• To develop a method for capturing individual-specific brain connectivity patterns.

Main Methods:

• Utilized a deep neural network (DNN) trained on Human Connectome Project rfMRI data.
• Learned features from time-varying functional connectivity (FC) for individual identification.
• Proposed the "fingerprint of FC" (fpFC) using nonlinear hidden layers.

Main Results:

• Successfully identified individuals with low error rates (2.9% for 300 individuals, 6.7% for 870 individuals).
• fpFCs demonstrated both common and individual-specific FC edges across various time windows.
• Validated model utility on an independent dataset, confirming transfer learning feasibility.

Conclusions:

• DNNs can effectively identify individuals based on rfMRI functional connectivity.
• The "fingerprint of FC" (fpFC) represents a novel approach to characterizing individual brain uniqueness.
• This technique offers insights into intrinsic brain modes and has potential for clinical applications.