Enhancing the network specific individual characteristics in rs-fMRI functional connectivity by dictionary learning

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Brain Connectivity

Background:

• Most fMRI studies analyze group data, overlooking individual brain variability.
• Individual connectomes, or unique brain connectivity patterns, are gaining research interest for participant identification.
• Subject-specific functional connectivity (FC) components hold potential for distinguishing individuals across testing sessions.

Purpose of the Study:

• To compare four dictionary-learning algorithms for extracting individual variability from network-specific FC.
• To assess the impact of Fisher Z and degree normalization on subject-specific components.
• To introduce a new metric, alongside differential identifiability, for evaluating extracted components.

Main Methods:

• Utilized resting-state functional Magnetic Resonance Imaging (rs-fMRI) data with 10 scans per subject.
• Applied four dictionary-learning algorithms to compute individual variability from network-specific FC.
• Compared Fisher Z and degree normalization techniques for functional connectivity.
• Introduced a novel metric and used differential identifiability to quantitatively evaluate subject-specific components.

Main Results:

• Common Orthogonal Basis Extraction (COBE) demonstrated strong performance in extracting subject-specific components.
• Fisher Z normalization, combined with COBE, yielded the most identifiable features for participants.
• The fronto-parietal and default mode networks showed significant individual-specific information.

Conclusions:

• Dictionary learning, particularly COBE with Fisher Z normalization, effectively captures individual brain connectivity.
• Subject-specific network features from rs-fMRI can reliably identify participants.
• This approach enhances the potential of neuroimaging for personalized neuroscience and subject tracking.