Breadcrumb
Method for Phase Contrast Based Magnetic Resonance Image Reconstruction
Benefit
- Reduces Scan Time: Achieves high‑quality reconstructions from highly undersampled data, enabling faster MRI exams.
- No Training Data Needed: Fully unsupervised method removes dependence on large, fully sampled datasets.
- Improved Image Quality: Provides higher SNR and more accurate phase maps across elastography, flow imaging, and thermometry.
- Physics‑Informed Reconstruction: Integrates vibration, flow, and thermal dynamics to increase reconstruction stability and accuracy.
- Versatile Across Modalities: Applicable to MRE, 4D flow, thermometry, and other phase‑contrast MRI techniques.
- Supports Clinical Decision‑Making: Enables more accurate quantitative biomarkers for neurological and systemic conditions.
- Easy Integration: Compatible with current MRI infrastructure and deployable on-scanner, on-premise, or via cloud computing.
Technology Description
This invention introduces a deep generative neural‑network reconstruction method that enables high‑quality phase‑contrast MRI from highly undersampled data, significantly reducing scan time for applications such as MR elastography, 4D flow MRI, and MR thermometry.
The method uses an unsupervised deep generative model that eliminates the need for large training datasets. It incorporates anatomical consistency constraints to preserve shared structural information across repeated acquisitions and applies physics‑based priors reflecting vibration, flow, and temperature dynamics. Through an iterative optimization process, the system jointly updates network parameters and acquisition‑specific input vectors to produce high‑resolution, high‑SNR images. This approach can be deployed on existing MRI scanners, clinical servers, or cloud environments.
UIRF Case No. 2025-025
Stage of Development
Initial experiment and proof of concept validation
IP Status: Patent Pending: Patent pending
Lead Researcher(s)
- Xi Peng, Department of Radiology, College of Medicine