Related Experiment Video
Updated: Sep 13, 2025
10:25
Brain Infarct Segmentation and Registration on MRI or CT for Lesion-symptom Mapping
Published on: September 25, 2019
48.4K
A Trust-Guided Approach to MR Image Reconstruction With Side Information
IEEE Transactions on Medical Imaging
|July 31, 2025
Summary
This study introduces the Trust-Guided Variational Network (TGVN) to accelerate MRI scans by using auxiliary data. TGVN enhances image quality and preserves details, significantly reducing scan times and costs.
Area of Science:
- Medical imaging and computational science.
- Development of advanced deep learning algorithms for magnetic resonance imaging (MRI).
Background:
- MRI scan acceleration is crucial for improving patient care and reducing healthcare expenses.
- Reconstructing diagnostic-quality images from sparse k-space data involves solving ill-posed linear inverse problems (LIPs).
- Prior knowledge, such as regularization or auxiliary data (side information), is essential for addressing ambiguities in LIPs.
Purpose of the Study:
- To present the Trust-Guided Variational Network (TGVN), an end-to-end deep learning framework for integrating side information into LIPs.
- To demonstrate TGVN's effectiveness in multi-coil, multi-contrast MRI reconstruction using under-sampled data.
Main Methods:
- Developed an end-to-end deep learning framework, TGVN, to integrate auxiliary data (side information) into MRI reconstruction.
- Utilized incomplete or low-SNR measurements from one MRI contrast as side information to reconstruct high-quality images of another contrast from under-sampled data.
- Applied TGVN to multi-coil, multi-contrast MRI reconstruction tasks.
Main Results:
- TGVN effectively and reliably integrates side information into linear inverse problems for MRI reconstruction.
- Achieved superior image quality compared to baseline methods that utilize side information.
- Preserved subtle pathological features even at high acceleration levels, demonstrating robustness across different contrasts, anatomies, and field strengths.
- Significantly reduced acquisition time while minimizing image artifacts (hallucinations).
Conclusions:
- TGVN offers a robust and effective deep learning solution for accelerating MRI acquisition.
- The framework successfully leverages auxiliary data to improve image quality and preserve diagnostic information.
- TGVN has the potential to drastically speed up MRI scans, leading to improved patient care and lower healthcare costs.
