Concurrent Modeling of Naturalistic Functional Brain Networks: A Four-Dimensional Multi-Pattern Spatio-temporal...

Area of Science:

• Neuroscience
• Computational Neuroscience
• Medical Imaging

Background:

• Understanding brain function relies on modeling whole-brain functional networks (FBNs) using fMRI.
• Existing shallow and deep models struggle to concurrently extract multiple FBNs and fully utilize 4D fMRI data features.

Purpose of the Study:

• To propose a novel Multi-Pattern Spatiotemporal Hybrid Attention 4D CNN (MSTHA-4DCNN) model.
• To concurrently capture spatiotemporal patterns of multiple FBNs by leveraging 4D fMRI data characteristics.

Main Methods:

• The MSTHA-4DCNN model integrates Multi-Pattern Spatial Attention 4D CNN (MSA-4DCNN) for spatial pattern extraction.
• Multi-Pattern Temporal Guided Attention Network (MT-GANet) models temporal representations guided by spatial patterns.
• The model was trained on a naturalistic fMRI dataset and validated on the Cam-CAN dataset.

Main Results:

• MSTHA-4DCNN demonstrated promising performance and generalization capabilities.
• The model effectively identified spatiotemporal patterns of FBNs concurrently.
• MSTHA-4DCNN outperformed existing state-of-the-art models in FBN identification.

Conclusions:

• MSTHA-4DCNN offers a potent tool for analyzing complex neural processes.
• The model advances the concurrent extraction of multiple FBNs from 4D fMRI data.
• This approach enhances our understanding of brain function through improved spatiotemporal network analysis.