MRI Data Reconstruction Using Network-Specific Weight Coefficients
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image reconstruction methods using machine learning and Data Consistency (DC) processes in magnetic resonance imaging face challenges in obtaining proper output results due to low learning efficiency when weight coefficients are shared among neural networks.
Innovation Solution
A data processing apparatus that generates and corrects partial sampling data using neural networks, enhancing learning efficiency by independently determining weight coefficients through an end-to-end training process with a Data Consistency (DC) process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If weight coefficients are shared among multiple neural networks in the iterative reconstruction process, then the computational complexity is reduced, but the learning efficiency deteriorates and proper output results cannot be obtained
Solution Approach 1:
The patent divides the weight coefficients into network-specific parameters, where each neural network has its own dedicated weight coefficients rather than sharing common weights. This segmentation allows each network to independently learn optimal parameters for its specific function in the iterative reconstruction process, resolving the contradiction between computational efficiency and learning effectiveness.
Solution Approach 2:
The patent implements parameter changes by allowing each neural network to have independently determined weight coefficients that are specific to each network. This parameter differentiation enables each network to adapt to its specific role in the reconstruction pipeline, improving learning efficiency while maintaining manageable computational complexity through structured parameter organization.
2Reliability
If end-to-end training is performed with independent weight coefficients for each neural network, then proper output results are obtained, but learning efficiency decreases
Solution Approach 1:
The patent applies preliminary action by pre-determining and fixing the weight coefficients for each neural network before the main iterative reconstruction process. This allows the networks to be trained independently with their optimal parameters established in advance, ensuring proper output results while avoiding the computational burden of end-to-end training with independent weight determination during the reconstruction process.
3Device complexity
If multiple neural networks are used in an iterative process with shared weight coefficients, then the reconstruction process can be simplified, but the ability to obtain proper output results deteriorates
Solution Approach 1:
The patent implements local quality by assigning network-specific weight coefficients to each neural network, allowing each network to have optimized parameters tailored to its specific function in the reconstruction pipeline. This local parameter optimization ensures that each network contributes effectively to the overall reconstruction accuracy while maintaining a simplified and manageable process architecture.
Data Source
AI summary
A data processing apparatus according to an embodiment includes processing circuitry. The processing circuitry is configured to output first complementary data by inputting, to a first neural network, first partial sampling data resulting from performing a partial sampling process; to obtain first corrected data, by performing a process to improve a consistency degree between the first complementary data and the first partial sampling data; to generate second partial sampling data, on the basis of the first corrected data and the first partial sampling data; and to output second complementary data, by inputting the second partial sampling data to a second neural network.