Related Experiment Video
Updated: Jan 8, 2026
Concurrent EEG and Functional MRI Recording and Integration Analysis for Dynamic Cortical Activity Imaging
Published on: June 30, 2018
A dense recurrent unrolling network leveraging spatio-temporal priors for highly-accelerated dynamic MRI
Bin Wang1, Yusheng Lian2, Wan Zhang2
1School of Printing and Packaging Engineering, Beijing Institute of Graphic Communication, Beijing 102600, China; Center for Metrology Scientific Data, National Institute of Metrology, Beijing 100029, China; Key Laboratory of Metrology Digitalization and Digital Metrology, State Administration for Market Regulation, Beijing 100029, China.
This study introduces a new deep learning method for faster dynamic magnetic resonance imaging (MRI) reconstruction. By improving temporal modeling and feature sharing, it enhances image quality and stability, even with significant undersampling.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Signal Processing
Background:
- Dynamic MRI requires fast, accurate reconstruction from undersampled data for high temporal resolution.
- Deep unrolling networks combine physics and learned priors but struggle with temporal relationships.
- Existing methods often process stages independently, limiting feature interaction and performance at high acceleration factors.
Purpose of the Study:
- To improve temporal prior learning in deep unrolling networks for dynamic MRI.
- To enhance reconstruction accuracy and temporal fidelity in highly accelerated scans.
- To develop a method that better exploits temporal dependencies and multi-stage collaboration.
Main Methods:
- Introduced a bidirectional recurrent convolutional unit for enhanced temporal dependency modeling.
- Incorporated inter-stage feature transmission to improve multi-stage collaboration.
- Evaluated the approach on dynamic MRI datasets with acceleration factors of 6×, 12×, and 24×.
Main Results:
- The proposed method consistently outperformed state-of-the-art unrolling and deep learning strategies.
- Demonstrated superior reconstruction accuracy and temporal fidelity under high undersampling conditions.
- Ablation studies confirmed the effectiveness of recurrent temporal learning and inter-stage feature transmission.
Conclusions:
- The novel deep unrolling approach significantly improves dynamic MRI reconstruction quality.
- Enhanced temporal modeling and feature sharing lead to better performance in accelerated scenarios.
- The method offers a promising solution for achieving high temporal resolution in clinical MRI.
