Multi-Slice MRI Reconstruction Using CNN Regularization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional MRI techniques face challenges in efficiently reconstructing images from multiple slices acquired using simultaneous multi-slice protocols, particularly in terms of reducing scan time and minimizing artifacts.
Innovation Solution
A method involving iterative optimization with a regularization operation implemented by multiple layers of a convolutional neural network, combined with a data-consistency operation, is used to reconstruct MRI images from undersampled datasets acquired using a simultaneous multi-slice protocol.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If simultaneous multi-slice protocol is used to reduce scan time, then scanning speed is improved, but image quality deteriorates due to noise amplification and artifacts
Solution Approach 1:
A convolutional neural network is introduced as an intermediary component between the undersampled k-space data and the final image reconstruction. The neural network learns the mapping relationship from compressed sensing measurements to high-quality images, effectively mediating the contradiction between fast scanning and image quality by denoising and artifact reduction through learned representations
Solution Approach 2:
The patent changes the reconstruction parameter from conventional iterative methods to deep learning-based reconstruction. By training a convolutional neural network on the relationship between undersampled k-space data and high-quality images, the system achieves fast reconstruction without noise amplification and artifacts, resolving the contradiction between scanning speed and image quality
2Manufacturing precision
If conventional iterative reconstruction methods are used, then image quality is maintained, but reconstruction time increases significantly
Solution Approach 1:
The patent replaces the mechanical iterative optimization process with a neural network-based reconstruction system. Instead of performing repeated iterative calculations to converge on an image solution, the trained neural network directly maps undersampled k-space data to high-quality images in a single forward pass, substituting the computational mechanical process with a learned nonlinear transformation
3Manufacturing precision
If deep learning reconstruction is applied, then noise and artifacts are reduced, but computational complexity increases
Solution Approach 1:
The convolutional neural network is pre-trained offline using a large dataset of paired undersampled k-space data and high-quality reference images. This preliminary training action captures the complex mapping relationships in the network weights, allowing the trained model to perform fast inference during actual imaging without requiring complex real-time computations, thus reducing online computational complexity
Data Source
AI summary
Various techniques of reconstructing multiple Magnetic Resonance Imaging, MRI, images for multiple slices based on an MRI measurement dataset that is acquired using a simultaneous multi-slice protocol and undersampling and K-space are disclosed. A convolutional neural network can be used to implement a regularization operation of an iterative optimization for the reconstruction, i.e., an unrolled neural network or variational neural network. A combination with Dixon imaging, i.e., separation of multiple chemical species, is disclosed.