A dynamic spatiotemporal representation framework for deciphering personal brain function.

Area of Science:

• Neuroscience
• Neuroimaging
• Computational Biology

Background:

• Functional magnetic resonance imaging (fMRI) provides insights into spontaneous human brain activity in vivo.
• The inherent complexity of fMRI signals poses challenges for creating accurate brain functional representations.
• Existing methods like amplitude of low-frequency fluctuations and Pearson's correlation are limited in capturing individual brain specifics.

Purpose of the Study:

• To introduce a novel state decomposition method for reducing fMRI signal complexity.
• To develop state-based metrics for quantifying brain activity and network interactions.
• To establish more discriminative and reproducible 'brain fingerprints' for individual identification and brain-phenotype modeling.

Main Methods:

• Brain dynamics are captured using temporal first-order derivatives.
• Spatial division into 'state sets' based on velocity and direction of signal change at each time point.
• Transformation of fMRI signals into discrete series of four fundamental states for encoding individual-specific information.

Main Results:

• State-based representations serve as more discriminative 'brain fingerprints' compared to conventional methods.
• Reproducible spatial patterns were observed across a heterogeneous cohort (n = 1015).
• Decoding of personal phenotypes (age, gender) from regional representations achieved high accuracy.
• State series equivalence outperformed existing network representations in predicting fluid intelligence.

Conclusions:

• The state decomposition framework effectively reduces fMRI signal complexity and enhances individual brain functional representations.
• This method provides a foundation for richer brain functional repertoire and improved brain-phenotype modeling.
• The developed state-based metrics offer a powerful tool for neuroscience research and clinical applications.