MRI Image Acquisition With Neural K-Space Motion Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing motion correction techniques in Magnetic Resonance Imaging (MRI) are either too time-consuming for clinical use or lack universal applicability across different imaging sequences, and they struggle to accurately correct for patient motion during the acquisition process.
Innovation Solution
A method using a trained machine learning model, specifically a neural network, to estimate pose parameters from additional k-space lines during the imaging protocol, allowing for real-time prospective motion correction without the need for scout image reconstruction, and compatible with any imaging sequence.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If retrospective motion correction techniques are used to correct motion artifacts after data acquisition, then motion artifacts are reduced, but reconstruction time increases to an unacceptable level for clinical routine
Solution Approach 1:
The patent applies preliminary action by acquiring additional k-space lines at the beginning of the imaging protocol before the actual imaging data acquisition. These preliminary k-space lines are used by the neural network to estimate pose parameters in advance, enabling fast prospective motion correction during the imaging protocol without delaying the reconstruction process.
Solution Approach 2:
The patent replaces the traditional mechanical optimization-based motion correction system with a neural network-based system. Instead of using iterative optimization algorithms that require significant computation time, the neural network directly estimates pose parameters from the additional k-space lines, dramatically reducing reconstruction time while maintaining motion correction quality.
2Measurement precision
If traditional motion correction methods are applied, then motion parameters can be estimated, but the methods lack universal applicability across different imaging sequences
Solution Approach 1:
The patent achieves universality by designing a neural network that can estimate pose parameters across any imaging sequence without requiring sequence-specific customization. The network is trained on diverse imaging data and can process additional k-space lines from different sequence types (e.g., multi-shot, single-shot, 2D, 3D) using the same architecture and processing approach, making the motion correction method universally applicable.
Solution Approach 2:
The patent applies parameter changes by using the neural network to directly output pose parameters (translation and rotation values) that are then fed into the motion correction process. The network learns to map from raw k-space line data to meaningful pose parameters through training, allowing it to adapt to different imaging sequences by learning sequence-specific patterns during training while maintaining the same core estimation mechanism.
3Measurement precision
If scout image reconstruction is used for motion estimation, then motion information can be derived, but the process becomes too time-consuming for real-time prospective correction
Solution Approach 1:
The patent extracts only the essential motion information directly from additional k-space lines without performing full scout image reconstruction. The neural network processes the raw k-space line data to directly estimate pose parameters, extracting motion information in its most compact and computationally efficient form, thereby eliminating the time-consuming intermediate step of reconstructing scout images.
Solution Approach 2:
The patent uses additional k-space lines that are acquired during the imaging protocol itself as a substitute (copy) for traditional scout images. These additional k-space lines contain sufficient information for motion estimation and can be processed much faster by the neural network, providing both motion information accuracy and computational efficiency.
4Measurement precision
If additional hardware or separate navigator scans are used for prospective motion correction, then accurate tracking of subject motion is achieved, but the acquisition time is prolonged
Solution Approach 1:
The patent merges the motion estimation function with the existing imaging protocol by incorporating additional k-space line acquisitions into the standard imaging sequence. Instead of using separate navigator scans or additional hardware systems, the motion estimation data is collected as an integrated part of the imaging protocol itself, eliminating extra acquisition time while maintaining accurate motion tracking through the neural network processing.
Data Source
AI summary
A method for acquiring a magnetic resonance image dataset of a body part of a subject includes acquiring a low-resolution magnetic resonance image of the body part, optionally acquiring a reference set of additional k-space lines within a central region of k-space, and acquiring further sets of additional k-space lines within a central region of k-space at intervals throughout the imaging protocol. The method includes applying a trained machine learning model to a dataset including the low-resolution image in k-space notation and a further set of additional k-space lines. A set of pose parameters is generated. The pose parameters are used for prospective and/or retrospective motion correction of the magnetic resonance image dataset.