Scepter: Weakly Supervised Framework for Spatiotemporal Dense Prediction of 4D Dynamic Brain Networks.

Area of Science:

• Neuroscience
• Computer Vision
• Medical Imaging

Background:

• Spatiotemporal brain dynamism involves complex neural activity patterns across space and time.
• Capturing these dynamics is challenging due to intricate neural interactions and computational demands.
• Existing brain parcellation methods have significant drawbacks.

Purpose of the Study:

• To develop a novel framework for weakly supervised spatiotemporal dense prediction of dynamic brain networks.
• To encode spatiotemporal characteristics of functional magnetic resonance imaging (fMRI) data for dynamic network prediction.
• To address limitations in current methods for analyzing brain dynamics.

Main Methods:

• Harnessed computer vision advances to frame the problem as spatiotemporal dense prediction.
• Developed a novel framework using an isotropic model architecture with ConvMixer modules.
• Introduced a weak supervision strategy using generated prior information due to lack of benchmarks and costly annotation.

Main Results:

• The framework successfully generated plausible brain network maps.
• The generated maps exhibit high dynamism and temporal variation.
• Results are consistent with established findings in brain dynamics research.

Conclusions:

• The proposed method offers a significant advancement in generating dynamic brain maps from fMRI data.
• This approach facilitates new perspectives on the complexity of brain function.
• The framework represents a paradigm shift in neuroscience research for understanding brain dynamism.