Deep Learning Network for Quantitative MRI Artifact Suppression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Quantitative magnetic resonance imaging (MRI) faces challenges with longer acquisition times and computational expenses due to undersampling artifacts and the need for iterative algorithms, particularly in fat quantification for diagnosing conditions like non-alcoholic fatty liver disease, which also suffers from motion artifacts and invasive biopsy procedures.
Innovation Solution
A two-stage deep learning network is employed to suppress undersampling artifacts and generate accurate quantitative parameter maps, including proton-density fat fraction (PDFF) and R2* maps, while providing uncertainty estimation to detect potential errors, thereby reducing acquisition and computational times.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If data undersampling is used to accelerate quantitative MRI acquisition, then acquisition time is reduced, but undersampling artifacts obscure image features and impact quantification accuracy
Solution Approach 1:
A deep learning network is introduced as an intermediary between undersampled k-space data and quantitative parameter maps. The network learns to map undersampled data to accurate quantitative maps by training on pairs of undersampled input data and reference quantitative maps generated from fully-sampled data, thereby eliminating artifacts while preserving quantification accuracy
Solution Approach 2:
The patent transforms the problem from direct image reconstruction to learning a mapping function between input parameters (undersampled k-space data) and output parameters (quantitative parameter maps). By changing the approach from artifact removal to direct parameter prediction, the system achieves both speed and accuracy
2Manufacturing precision
If constrained reconstruction methods (e.g., compressed sensing) are used to reduce undersampling artifacts, then image quality is improved, but iterative algorithms increase computational time
Solution Approach 1:
The patent replaces iterative optimization algorithms (mechanical/computational process) with a trained deep learning network (statistical model). The network performs artifact suppression and quantification in a single forward pass rather than through multiple iterative refinements, dramatically reducing computational time while maintaining image quality
Solution Approach 2:
The deep learning network is pre-trained on large datasets of paired undersampled and reference images. This preliminary training phase allows the network to learn optimal artifact suppression strategies, which are then applied rapidly during actual quantitative MRI scans without requiring iterative computation
3Manufacturing precision
If multi-echo 3D Cartesian sequence is used for liver fat quantification, then quantification accuracy is improved, but breath-holding requirement limits volumetric coverage and resolution
Solution Approach 1:
The patent inverts the conventional approach by using non-Cartesian radial sampling with self-gating instead of Cartesian sampling with breath-holding. This reversal allows free-breathing acquisition while maintaining motion robustness through self-gated artifact suppression, improving patient comfort without sacrificing quantification accuracy
4Ease of operation
If self-gated free-breathing radial data acquisition is used, then motion robustness is improved, but radial undersampling artifacts degrade image quality and quantification accuracy
Solution Approach 1:
A deep learning network serves as an intermediary that processes self-gated radial data to eliminate undersampling artifacts. The network is trained specifically on self-gated data to learn the characteristic artifact patterns and their removal, preserving both motion robustness and quantification accuracy
Solution Approach 2:
The patent applies local quality enhancement by using self-gating to select data from specific respiratory phases (end-expiration) while suppressing artifacts in other regions. The deep learning network further refines this by applying localized artifact suppression to preserve image quality in critical regions while maintaining overall motion robustness
Data Source
AI summary
A method for generating magnetic resonance imaging (MRI) quantitative parameter maps includes receiving at least one multi-contrast magnetic resonance (MR) image of a subject, providing the image to an artifact suppression deep learning network of a two-stage deep learning network and generating at least one multi-contrast MR image with suppressed undersampling artifacts using the artifact suppression deep learning network. The method further includes providing the at least one multi-contrast MR image with suppressed undersampling artifacts to a parameter mapping deep learning network of the two-stage deep learning network, generating at least one quantitative MR parameter map and generating an uncertainty estimation map for the at least one quantitative MR parameter map using the parameter mapping deep learning network. The method further includes displaying at least one multicontrast MR image with suppressed undersampling artifacts, at least one quantitative MR parameter map, and the corresponding uncertainty estimation map on a display.