4D Modeling of fMRI Data via Spatio-Temporal Convolutional Neural Networks (ST-CNN).

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Computational Neuroscience

Background:

• Functional Magnetic Resonance Imaging (fMRI) enables brain mechanism investigation.
• Analyzing 4D fMRI data for joint spatial-temporal patterns remains challenging.
• Existing methods often analyze spatial or temporal domains separately, neglecting fMRI's 4D nature.

Purpose of the Study:

• To propose a novel framework for simultaneous modeling of spatial and temporal patterns in 4D fMRI data.
• To automatically identify functional brain networks by jointly extracting their characteristics.
• To address the methodological gap in effectively investigating the 4D nature of fMRI data.

Main Methods:

• Development of a novel spatio-temporal convolutional neural network (ST-CNN).
• Joint extraction of spatial and temporal characteristics from fMRI data.
• Application of the ST-CNN framework for identifying the Default Mode Network (DMN).

Main Results:

• The ST-CNN framework demonstrates sufficient generalizability, identifying the DMN from diverse datasets (different cognitive tasks and resting states) after training on a single dataset.
• The joint-learning scheme effectively captures intrinsic relationships between spatial and temporal characteristics of the DMN.
• Accurate identification of the DMN from independent datasets was achieved.

Conclusions:

• The ST-CNN model offers a powerful new approach for analyzing fMRI data by jointly considering spatial and temporal information.
• This framework provides new tools and insights for fMRI analysis in cognitive and clinical neuroscience.
• The model's generalizability highlights its potential for broad application in brain network research.