Multi-Kernel GRAPPA EPI Reconstruction for MRI Ghosting Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Echo planar imaging (EPI) in MRI is severely limited by hardware system imperfections such as gradient errors and field perturbations, leading to prominent image artifacts, especially in advanced imaging practices with higher acquisition acceleration, stronger gradients, and higher field strengths, which hinder the improvement of image performance and speed.
Innovation Solution
A generalized multi-kernel GRAPPA technique using machine learning to correct spatial-varying field imperfections, employing multi-layer perceptron (MLP) networks to train kernels for artifact-free image reconstruction without increasing scan time, applicable across different systems and field strengths.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If dual polarity averaging (DPA) is used to correct field imperfections, then image quality improves, but acquisition time doubles
Solution Approach 1:
The patent segments the correction process into two phases: (1) acquire calibration data in dual polarity during a separate preparation step, and (2) use the learned correction kernels to correct single polarity clinical data in real-time. This segmentation allows the time-consuming dual polarity acquisition to be performed only once for calibration, while clinical scans use efficient single polarity acquisition with post-processing correction.
Solution Approach 2:
The patent performs preliminary action by acquiring dual polarity calibration data and training correction kernels before actual clinical scanning. The correction kernels are pre-computed from the calibration data, enabling fast correction of single polarity clinical data without requiring dual polarity acquisition during the actual scan, thus improving clinical workflow efficiency.
2Productivity
If single polarity EPI acquisition is used to maintain scanning efficiency, then productivity improves, but field imperfection artifacts worsen
Solution Approach 1:
The patent introduces correction kernels as an intermediary element that mediates between the single polarity acquired data and the desired corrected image. These kernels, learned from dual polarity calibration data, act as a transformation operator that removes field imperfection artifacts from single polarity clinical data, enabling efficient single polarity scanning without sacrificing image quality.
Solution Approach 2:
The patent replaces the mechanical dual polarity acquisition system with a computational correction system. Instead of physically acquiring dual polarity data during every clinical scan, the system uses machine learning-based kernel correction to achieve similar artifact suppression, substituting hardware-based dual polarity scanning with software-based post-processing correction.
3Productivity
If higher gradient strength and slew rate are used to improve imaging speed, then productivity improves, but eddy current effects and spatial variation of field imperfection worsen
Solution Approach 1:
The patent changes the parameter representation of field imperfections from fixed models to spatially-varying models that adapt to different gradient conditions. The correction kernels are trained to capture spatial variations in field imperfections that arise from high gradient strength and slew rate, enabling accurate correction even when imaging parameters are pushed to higher speeds.
Data Source
AI summary
A method for magnetic resonance imaging (MRI) includes: performing with an MRI scanner, an MRI data acquisition to acquire MRI data including calibration data for kernel training and regular EPI data in single polarity; performing preprocessing of the MRI data including performing readout interpolation to Cartesian grids and first-order gradient delay correction; performing a generalized multi-kernel GRAPPA using network-based kernels trained from the calibration data to correct spatial-varying field imperfections; applying the network-based kernels to the regular EPI data to obtain clean data; and reconstructing corrected single-polarity EPI data to obtain clean images using a parallel imaging reconstruction algorithm.