MRI Reconstruction With Statistical Regularization and K-Space Integrity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Magnetic resonance image reconstruction based on undersampled k-space data is an ill-posed problem, leading to inaccurate and unstable reconstructions due to high neural network freedom and lack of shared parameters, resulting in potential divergence from ground truth images.
Innovation Solution
A method involving a regularization process using a first neural network to correct pixel value statistics, followed by a data integrity process to align k-space data, enhancing stability and accuracy by iteratively refining image data to approximate ground truth.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If undersampled k-space data is used for magnetic resonance image reconstruction, then scanning time is reduced, but reconstruction accuracy deteriorates due to insufficient information
Solution Approach 1:
The patent implements an iterative reconstruction process where the neural network continuously refines the image estimate by comparing reconstructed images with the undersampled k-space data. The network adjusts its parameters based on the discrepancy between the current reconstruction and the actual measured data, progressively improving accuracy while maintaining reduced scanning time.
Solution Approach 2:
The patent employs a neural network with learnable parameters that are optimized during training to map undersampled k-space data to high-quality images. The network transforms the ill-posed reconstruction problem into a parameter estimation problem, where the parameters are tuned to capture the underlying image structure and compensate for the missing k-space information.
2Measurement precision
If neural networks with high freedom for parameters are used for reconstruction, then reconstruction precision improves, but stability deteriorates due to divergence from ground truth
Solution Approach 1:
The patent incorporates a feedback mechanism where the neural network's output is continuously evaluated against the undersampled k-space data. The network parameters are adjusted based on the reconstruction error, preventing divergence from ground truth while maintaining high precision. This feedback loop ensures that the high parameter freedom of the network is guided toward physically meaningful solutions.
Solution Approach 2:
The patent performs preliminary training of the neural network on simulated data before实际应用. During training, the network learns the relationship between undersampled k-space data and ground truth images, establishing a foundation that guides the network's high parameter freedom toward stable and reliable reconstructions. This preliminary action prevents divergence during actual reconstruction tasks.
3Device complexity
If no processing is performed on input data or output data of the image space regularization block, then device complexity is reduced, but reconstruction quality deteriorates
Solution Approach 1:
The patent applies parameter changes through the neural network's learnable parameters, which transform the input image data and produce improved output. Instead of complex structural modifications, the network uses optimized parameters to capture image characteristics and enhance reconstruction quality, achieving high performance with relatively simple processing architecture.
Data Source
AI summary
A magnetic resonance image reconstruction method according to an embodiment is a magnetic resonance image reconstruction method for reconstructing magnetic resonance image data on the basis of undersampled k-space data and includes: generating second image data by performing a regularization process within an image space while using a first neural network on first image data generated on the basis of the undersampled k-space data; generating third image data by correcting the second image data so that a pixel value statistical feature of the second image data approximates a pixel value statistical feature of the first image data; and generating fourth image data by performing a data integrity process on the third image data so that k-space data corresponding to the third image data approximates the undersampled k-space data.