Spatiotemporal trajectories in resting-state FMRI revealed by convolutional variational autoencoder

Area of Science:

• Neuroscience
• Computational Neuroscience
• Functional Neuroimaging

Background:

• Resting-state functional magnetic resonance imaging (rs-fMRI) studies reveal dynamic changes in brain functional connectivity over time.
• Existing methods often simplify these dynamics by modeling brain activity as transitions between discrete states.

Purpose of the Study:

• To develop a more accurate model of brain activity dynamics in resting-state fMRI.
• To identify fundamental spatiotemporal patterns underlying continuous brain activity.

Main Methods:

• A variational autoencoder was trained on resting-state fMRI data.
• The spatiotemporal features of the latent variables from the trained network were evaluated.

Main Results:

• A small set of approximately orthogonal whole-brain spatiotemporal patterns were identified.
• These patterns effectively capture the prominent features of rs-fMRI data.
• The identified patterns serve as building blocks for constructing complex spatiotemporal dynamics.

Conclusions:

• Brain activity in resting state is better represented as a continuous process rather than discrete states.
• The discovered spatiotemporal patterns offer insights into brain network structures and functional connectivity gradients.
• This approach provides a novel framework for understanding brain dynamics.