Deep Learning Motion Artifact Prediction in 3D MRI
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for assessing motion artifacts in magnetic resonance (MR) images are inadequate as they rely on 2D image scoring, which fails to accurately reflect three-dimensional patient motion, leading to inaccurate quality assessment and potential misdiagnosis.
Innovation Solution
A deep learning-based method that simulates three-dimensional patient motion and k-space line order to generate training data, allowing for the prediction of motion artifacts in 3D MR images, using a network architecture capable of handling anisotropic data and multiple motion scenarios.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If 2D image-based scoring methods are used to assess motion artifacts, then the assessment process is simple and fast, but the accuracy of motion artifact detection is insufficient because real patient motion is three-dimensional
Solution Approach 1:
The patent transitions from 2D image-based scoring to 3D volumetric MR data scoring. The system processes three-dimensional volumetric data instead of two-dimensional slices, adding the depth dimension to accurately capture three-dimensional patient motion artifacts. This dimensional expansion enables the scoring system to reflect the true nature of patient motion in 3D space, significantly improving detection accuracy.
Solution Approach 2:
The patent creates synthetic training data by copying and transforming ground truth 3D representations. Multiple copies of the original 3D data are generated with various motion transformations applied, creating a comprehensive training dataset without requiring additional physical scans. This copying approach enables robust model training while maintaining computational efficiency.
2Measurement precision
If volumetric 3D representation is used for motion assessment, then the accuracy of motion artifact evaluation is improved, but the computational complexity and processing time increase
Solution Approach 1:
The patent performs preliminary actions by pre-processing the volumetric MR data into a standardized format and pre-computing motion transformations during the training phase. The scoring model is trained in advance on extensively transformed 3D data, so that during actual assessment, the model can quickly evaluate new scans without performing computationally intensive transformations in real-time. This preliminary preparation significantly reduces processing time for clinical assessments.
Solution Approach 2:
The patent employs parameter changes by applying various motion transformation parameters (translation, rotation, scaling) to generate diverse training samples from a single ground truth 3D representation. By systematically varying these parameters, the system creates a comprehensive training dataset that covers the full range of possible patient motions, enabling the model to generalize well to new scans while maintaining efficient processing.
3Reliability
If realistic 3D motion simulation with k-space line order is performed to generate training data, then the training data quality is improved, but the data generation process becomes more complex and time-consuming
Solution Approach 1:
The patent uses copying to generate multiple versions of the ground truth 3D representation by applying different motion transformations. Instead of acquiring separate scans for each motion scenario, the system creates synthetic copies of the original data with various motion patterns applied, significantly accelerating training data generation while maintaining realism.
Solution Approach 2:
The patent replaces the mechanical process of acquiring multiple physical scans with different motion patterns by using computational simulation. Instead of physically moving the patient or scanner to create varied training data, the system uses software-based k-space line order manipulation and motion simulation to generate realistic training scenarios, dramatically improving productivity.
Data Source
Figure 1~3
Figure 4
Figure 5~8
AI summary
For determination of motion artifact in MR imaging, motion of the patient in three dimensions is used with a measurement k-space line order based on one or more actual imaging sequences to generate training data. The MR scan of the ground truth three-dimensional (3D) representation subjected to 3D motion is simulated using the realistic line order. The difference between the resulting reconstructed 3D representation and the ground truth 3D representation is used in machine-based deep learning to train a network to predict motion artifact or level given an input 3D representation from a scan of a patient. The architecture of the network may be defined to deal with anisotropic data from the MR scan.