Related Experiment Video
Updated: Oct 2, 2025
05:55
Modeling the Functional Network for Spatial Navigation in the Human Brain
Published on: October 13, 2023
1.2K
Attention module improves both performance and interpretability of four-dimensional functional magnetic resonance
Zhoufan Jiang1, Yanming Wang1, ChenWei Shi1
1Center for Biomedical Imaging, University of Science and Technology of China, Hefei, Anhui, China.
Human Brain Mapping
|February 25, 2022
Summary
This study introduces a novel 4D deep neural network (DNN) with attention modules for decoding brain cognitive states from fMRI data. The model achieves high accuracy and offers interpretable insights into brain activity patterns.
Area of Science:
- Neuroscience
- Machine Learning
- Neuroimaging
Background:
- Decoding brain cognitive states from neuroimaging signals is crucial in neuroscience.
- Deep neural networks (DNNs) show promise but lack interpretability.
- Interpreting DNNs is essential for understanding brain function.
Purpose of the Study:
- To develop an interpretable DNN for decoding brain cognitive states.
- To integrate attention modules and 4D convolution for enhanced fMRI analysis.
- To investigate the interpretability of DNNs in brain decoding.
Main Methods:
- Integrated attention modules into DNNs for brain decoders.
- Employed a four-dimensional (4D) convolution operation for fMRI temporo-spatial analysis.
- Utilized datasets from the Human Connectome Project (HCP) and BOLD5000.
Main Results:
- Achieved 97.4% accuracy on seven HCP task benchmarks, outperforming previous methods.
- Visualization revealed hierarchical, task-specific attention masks.
- Transfer learning demonstrated adaptive changes in high-level attention masks.
Conclusions:
- The proposed 4D DNN with attention facilitates interpretable brain state decoding.
- The model enhances understanding of DNNs in neuroscience research.
- Findings support the utility of attention mechanisms for neuroimaging data analysis.
