Adaptive Sparsifying Transform Learning for MRI Image Reconstruction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current imaging modalities, such as MRI and CT scanning, face challenges in reconstructing high-quality images from imperfect, incomplete, or degraded data, particularly due to non-adaptive sparsifying transforms and the need for extensive computational resources, leading to artifacts and increased radiation exposure.
Innovation Solution
The development of a transform learning method that adapts sparsifying transforms to the data, allowing for efficient reconstruction of images from limited measurements by combining physics and statistics-based models, and using alternating minimization algorithms to learn sparse representations and reconstruct images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If fixed, non-adaptive sparsifying transforms are used in compressed sensing MRI, then the reconstruction method is simpler and faster, but the reconstructions suffer from many artifacts at high undersampling factors
Solution Approach 1:
The patent applies dynamics by transitioning from fixed, non-adaptive sparsifying transforms to adaptive transforms that are learned from data. The transform is no longer static but dynamically adjusted based on the specific imaging task and data characteristics, allowing the system to adapt to different undersampling factors and minimize artifacts while maintaining computational efficiency through iterative learning processes.
2Productivity
If traditional Nyquist sampling is used, then the image reconstruction is accurate, but the imaging time is long and clinical throughput is low
Solution Approach 1:
The patent applies partial action by using compressed sensing to acquire only a subset of k-space data (partial sampling) rather than following traditional Nyquist sampling requirements. By combining this partial data acquisition with adaptive sparsifying transforms and iterative reconstruction algorithms, the system achieves both reduced imaging time (improved productivity) and maintained reconstruction accuracy through intelligent signal processing.
3Object-affected harmful factors
If low-dose imaging methods are used, then radiation exposure is reduced, but the data becomes incomplete and degraded requiring complex reconstruction
Solution Approach 1:
The patent applies feedback through iterative reconstruction algorithms that continuously refine the image estimate based on the degraded low-dose data. The adaptive sparsifying transforms provide feedback about the underlying signal structure, allowing the reconstruction algorithm to iteratively correct artifacts and improve image quality. This feedback mechanism enables the system to handle incomplete and degraded data while minimizing radiation exposure.
4Manufacturing precision
If fixed signal models like total-variation regularization are used, then the reconstruction is faster and more stable, but the images have patchy artifacts due to simplistic signal modeling
Solution Approach 1:
The patent applies dynamics by replacing fixed signal models with adaptive transforms that are learned from data specific to each imaging task. Instead of using a one-size-fits-all model like total-variation regularization, the system dynamically adapts the transform based on the actual image characteristics and acquisition parameters. This allows the system to capture complex signal structures more accurately (improved manufacturing precision) while maintaining computational efficiency through optimized learning processes.
Data Source
AI summary
A system executes efficient computational methods for high quality image reconstructions from a relatively small number of noisy (or degraded) sensor imaging measurements or scans. The system includes a processing device and instructions. The processing device executes the instructions to employ transform learning as a regularizer for solving inverse problems when reconstructing an image from the imaging measurements, the instructions executable to: adapt a transform model to a first set of image patches of a first set of images containing at least a first image, to model the first set of image patches as sparse in a transform domain while allowing deviation from perfect sparsity; reconstruct a second image by minimizing an optimization objective comprising a transform-based regularizer that employs the transform model, and a data fidelity term formed using the imaging measurements; and store the second image in the computer-readable medium, the second image displayable on a display device.