MR Image Reconstruction Network Parameter Refinement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine-learning-based image reconstruction methods in magnetic resonance (MR) imaging fail to achieve desired image quality and computational efficiency, as they either require a second image for blending, which is not available, or lose computational benefits in model-based iterative reconstruction.
Innovation Solution
A conditional unrolled neural network architecture with a refinement stage is used to learn and refine regularization parameters, ensuring desired image quality while maintaining computational efficiency by alternating between data-consistency and machine learning network blocks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If machine-learning-based image reconstruction is used to accelerate MR data acquisition, then scan speed is improved, but image quality deteriorates
Solution Approach 1:
The reconstruction process is divided into two distinct stages: a training stage where the neural network learns to reconstruct images from compressed sensing data, and an inference stage where the trained network rapidly reconstructs images from new acquired data. This segmentation allows the system to achieve both high speed during inference while maintaining image quality through the learning process.
Solution Approach 2:
A training stage is performed beforehand to pre-compute optimal network parameters by blending reference images with compressed sensing reconstructed images. This preliminary action stores the learned characteristics in the neural network, enabling rapid high-quality reconstruction during the actual scanning and inference phase without requiring time-consuming processing during the scan itself.
2Manufacturing precision
If network parameter values are refined to improve image quality, then image quality is improved, but computational complexity increases
Solution Approach 1:
The refinement of network parameter values is performed during a preliminary training stage before the actual image reconstruction is needed. By pre-computing and storing the refined parameters in the neural network, the system achieves high image quality during inference without incurring the computational complexity of parameter refinement during the actual scanning process.
Solution Approach 2:
The training process creates an optimized copy of the network parameters by blending the original network output with reference images. This copied and refined parameter set is then used during inference, achieving improved image quality while keeping the inference process computationally efficient since the heavy refinement work was already done during training.
Data Source
AI summary
A method for machine learning includes learning, during a training stage, network parameter values of a neural network to obtain a trained neural network configured to perform reconstruction of medical images; refining, during a subsequent refinement stage, the learned network parameter values to generate refined network parameter values defining a refined neural network; and applying input medical image data to the refined neural network to generate a reconstructed medical image. The method retains benefits of machine learning image reconstruction to obtain a desired reconstructed image.