MRI Image Recovery Using Score-Based Diffusion Sampling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing MRI reconstruction methods face challenges in accurately modeling the data distribution of magnetic resonance imaging (MRI) and require well-curated large databases for supervised learning, while score-based diffusion models offer high sample quality without adversarial training but lack effective application in MRI recovery.
Innovation Solution
A method using a score-based diffusion model to train a continuous time-dependent score function with denoising score matching, allowing sampling from a conditional distribution given measurements, and iteratively solving a stochastic differential equation (SDE) with data consistency to recover MRI images, even with complex-valued data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If iterative methods with hand-crafted priors are used for MRI reconstruction, then data consistency can be achieved, but the modeling accuracy of data distribution is insufficient
Solution Approach 1:
The patent replaces hand-crafted mechanical priors with a data-driven score-based diffusion model that learns the data distribution from actual MRI images. The score function ∇log p(x) is trained using denoising score matching, allowing the model to capture complex data distributions without manual prior construction. This substitution of mechanical prior design with automated learning-based prior modeling resolves the contradiction between modeling accuracy and construction difficulty.
2Manufacturing precision
If data-driven deep learning methods are used for MRI recovery, then image quality can be improved, but the method relies heavily on well-curated large databases which are hard to obtain
Solution Approach 1:
The patent employs self-supervised learning where the model trains on the MRI images themselves without requiring paired ground truth data or extensive curated databases. The score-based diffusion model learns the data distribution by analyzing the statistical properties of the input images, enabling the system to serve itself by extracting useful representations from the data alone. This self-service approach eliminates the dependency on large curated databases while maintaining high image recovery accuracy.
3Reliability
If score-based diffusion models are used for MRI recovery, then high sample quality can be achieved without adversarial training, but the application to MRI recovery is not yet effective
Solution Approach 1:
The patent develops a universal score-based diffusion framework that can be applied to various MRI recovery tasks including compressed sensing, parallel imaging, and image reconstruction. The score function trained on general MRI data distributions can handle different acquisition patterns and reconstruction requirements, making the model versatile across multiple MRI applications. This universality allows the same fundamental model to address diverse MRI recovery problems without requiring task-specific retraining, thereby resolving the contradiction between sample quality and applicability.
Data Source
AI summary
A magnetic resonance imaging (MRI) recovery method using a score-based diffusion model and an apparatus thereof are provided. The MRI recovery method using the score-based diffusion model, which is performed by a computer, is implemented, including training a continuous time-dependent score function with denoising score matching and sampling data from a conditional distribution given the measurements, leveraging the learned score function, and recovering an image.