MRI Dataset Acquisition with Neural K-Space Motion Tracking
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 universality across imaging sequences, and they often fail to fully correct rotation and spin-history artifacts.
Innovation Solution
A method using a trained machine learning model, particularly a neural network, estimates pose parameters from additional k-space lines during the imaging protocol, allowing for fast prospective motion correction that adapts the field of view in real-time, 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 significantly making it unacceptable for clinical routine
Solution Approach 1:
The patent applies preliminary action by acquiring navigator echoes at the beginning of each shot to capture motion information before the main imaging data acquisition. This allows motion parameters to be determined in advance, enabling prospective motion correction that adapts the field of view during the imaging protocol without delaying reconstruction
Solution Approach 2:
The patent extracts only the essential motion information from the navigator echoes - specifically the center of k-space data containing motion parameters - rather than processing the entire dataset. This extraction approach minimizes computational overhead and reconstruction time while maintaining motion correction effectiveness
2Reliability
If camera-based approaches or navigators are used for prospective motion correction, then motion tracking is achieved, but device complexity increases and acquisition time is prolonged
Solution Approach 1:
The patent employs self-service by using the MRI system's own navigator echoes and k-space data to track motion, rather than requiring external camera systems or separate navigator hardware. The existing MRI apparatus components perform dual functions: both imaging and motion tracking, eliminating the need for additional complex hardware
Solution Approach 2:
The patent achieves multi-functionality by designing the navigator echo acquisition to serve dual purposes: capturing motion information for prospective correction and providing reference data for retrospective correction. This universal approach allows a single system to perform multiple motion correction functions without adding device complexity
3Productivity
If deep learning techniques are applied to estimate subject motion from k-space data, then motion correction speed is improved, but the method requires special sequence features and cannot be universally applied to any imaging sequence
Solution Approach 1:
The patent achieves universality by using a machine learning model that processes standard navigator echo data and k-space information that are commonly available across different MRI sequences. The model is trained to recognize motion patterns from this universal data format, enabling application to various imaging sequences without requiring sequence-specific modifications
Solution Approach 2:
The patent applies parameter changes by adapting the machine learning model to work with different k-space sampling patterns and navigator echo configurations used in various MRI sequences. The model learns to extract motion parameters from different data representations, maintaining high correction speed across diverse sequence types
Data Source
Figure 1
Figure 2~3
Figure 4~5
AI summary
The invention relates to a method for acquiring a magnetic resonance image dataset of a body part of a subject (U), the method comprising the steps of: (a) acquiring a low-resolution magnetic resonance image (40) of the body part, (b1) optionally acquiring a reference set (38) of additional k-space lines within a central region (16) of k-space, (b2) acquiring further sets (36) of additional k-space lines (12) within a central region (16) of k-space at intervals throughout the imaging protocol, (c) applying a trained machine learning model (42) to a dataset (34) comprising the low-resolution image (40) in k-space notation and a further set of additional k-space lines (36), wherein a set of pose parameters (44) is generated, and (d) using the pose parameters (44) for prospective (46) and/or retrospective motion correction of the magnetic resonance image dataset.