MR Image Super-Resolution With Learned Gibbs Artifact Suppression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional super-resolution processing on magnetic resonance (MR) images in k-space leads to Gibbs artifacts due to zero padding, and filters used to reduce these artifacts diminish the effectiveness of the super-resolution process.
Innovation Solution
A medical image processing apparatus that uses a learned model to combine a first MR image reconstructed with super-resolution processing and a second MR image with suppressed artifacts, generating a third MR image with the same resolution and reduced artifacts, by applying different filter strengths and zero-filling techniques.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If zero padding is executed in the high-frequency region outside the MR data in the k-space to perform super-resolution processing, then the resolution of the MR image is improved, but Gibbs artifacts appear in the super-resolution MR image
Solution Approach 1:
The patent applies a filter to reduce the termination of end data in the k-space, which converts the harmful Gibbs artifacts into a controlled processing step. The filter processes the MR data to suppress artifacts while preserving the super-resolution effect, effectively turning the harmful artifact problem into a beneficial artifact suppression mechanism.
2Object-generated harmful factors
If a filter is applied to reduce the termination of end data in the MR data in the k-space to reduce Gibbs artifacts, then the Gibbs artifacts are reduced, but the end data is reduced and the effect of super-resolution is reduced
Solution Approach 1:
The patent changes the parameter of filter strength dynamically. It determines the strength of the filter based on the resolution of the MR image, using a higher resolution image for training to optimize the filter parameters. This allows the filter to adaptively balance artifact suppression with preservation of super-resolution effects.
3Measurement precision
If interpolation methods such as bicubic or Lanczos are executed on the MR image to perform super-resolution processing in the image space, then the resolution is improved, but artifacts similar to Gibbs artifacts appear in the super-resolution MR image
Solution Approach 1:
The patent replaces conventional interpolation methods (bicubic, Lanczos) with a deep learning-based super-resolution model. This substitution eliminates the need for traditional interpolation algorithms that generate artifacts, while achieving superior resolution through the learned model's optimization.
Data Source
AI summary
A medical image processing apparatus according to the present embodiment includes processing circuitry. The processing circuitry inputs a first magnetic resonance image reconstructed with super-resolution processing on magnetic resonance data and a second magnetic resonance image obtained by imaging the same object as that of the first magnetic resonance image and with artifacts suppressed compared with the first magnetic resonance image, to a leaned model, the learned model being configured to output a third magnetic resonance image having the same resolution as that of the first magnetic resonance image and with the artifacts suppressed, generates the third magnetic resonance image based on the first magnetic resonance image and the second magnetic resonance image, using the learned model.