Parallel MR Image Reconstruction Using Coil Maps to Reduce Aliasing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing MR image reconstruction methods in parallel imaging struggle with aliasing artifacts due to regular undersampling, particularly in conventional and MLM-based reconstructions, limiting the effectiveness of aliasing reduction and acceleration factors.
Innovation Solution
The method employs effective coil sensitivity maps generated using reconstruction weights to correct aliasing in position space, combining techniques like GRAPPA or CAIPIRINHA with machine learning models (MLMs) for enhanced image reconstruction, optimizing a loss function to improve aliasing correction and image quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If regular undersampling is used in parallel MR imaging to increase acceleration factor, then productivity is improved, but aliasing artifacts increase causing manufacturing precision to deteriorate
Solution Approach 1:
The method performs preliminary determination of coil sensitivity maps using fully sampled k-space data before the actual undersampled imaging. These pre-determined sensitivity maps are then used in the aliasing correction process during reconstruction, enabling effective artifact reduction even with high acceleration factors
Solution Approach 2:
The patent introduces coil sensitivity maps as an intermediary element that mediates between the undersampled k-space data and the final image reconstruction. These sensitivity maps serve as a bridge to separate and correct aliased signals, enabling accurate image recovery despite aggressive undersampling
2Device complexity
If conventional reconstruction methods are used with regular undersampling, then device complexity is kept low, but measurement precision deteriorates due to aliasing artifacts
Solution Approach 1:
The patent replaces traditional iterative optimization methods with a machine learning-based approach. A neural network model is trained to directly predict corrected image data from undersampled k-space inputs, substituting complex iterative algorithms with a trained model that achieves comparable or superior precision more efficiently
3Manufacturing precision
If machine learning models are used for image enhancement, then manufacturing precision is improved, but device complexity increases
Solution Approach 1:
The reconstruction process is segmented into distinct functional modules: a first neural network for initial image reconstruction from undersampled data, and a second neural network for enhancement and artifact correction. This segmentation allows each network to be optimized for its specific task while maintaining manageable overall system complexity
Solution Approach 2:
The patent employs a unified machine learning framework that performs multiple functions: the first network handles basic reconstruction, while the second network provides enhancement and aliasing correction. This multi-functional approach consolidates what could be separate processing stages into an integrated system
Data Source
Figure 1
Figure 2~4
Figure 5
AI summary
For image reconstruction in parallel MR imaging, a respective set of regularly undersampled MR measurement data in k-space representing an imaged object (6) is received for each of a plurality of coil channels. For each pair of coil channels of the plurality of coil channels, a respective set of reconstruction weights for reconstructing MR data at k-space points, which are not measured according to the undersampling, from the MR measurement data, is received. For each of the plurality of coil channels, a respective coil sensitivity map is determined depending on the respective sets of reconstruction weights for the respective coil channel. A reconstructed MR image is generated based on the coil sensitivity maps.