3D Spatiotemporal CNN for Phase-Contrast MRI Aliasing Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current phase-contrast MRI techniques face challenges in accurately measuring blood flow due to phase aliasing, particularly in cardiac 4D flow MRI, where velocity encoding parameters often result in artifactual reversals and low velocity-to-noise ratios, and existing algorithms fail to account for spatiotemporal correlations.
Innovation Solution
A deep learning-based phase unwrapping model using 3D convolutions and (2+1)D separated convolutions is developed to correct phase aliasing in phase-contrast MRI data, trained on simulated low velocity encoding values to efficiently generate corrected velocity maps, addressing the inefficiencies of existing methods.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If venc is increased to prevent aliasing, then aliasing is reduced, but velocity-to-noise ratio decreases
Solution Approach 1:
The patent applies preliminary action by training a deep learning model in advance on simulated phase-contrast MRI data with known velocity fields. The model learns to predict velocity-to-noise ratios and detect aliasing artifacts before actual measurement, enabling correction of aliased data without requiring increased venc settings that would degrade velocity-to-noise ratio.
2Device complexity
If 2D phase unwrapping algorithms are used, then computational simplicity is maintained, but spatiotemporal correlations are not accounted for
Solution Approach 1:
The patent transitions from 2D to 3D phase unwrapping by incorporating the temporal dimension. The deep learning model processes phase-contrast MRI data as 3D volumes (x, y, time), allowing the algorithm to exploit spatiotemporal correlations across multiple time points while maintaining computational efficiency through optimized neural network architectures.
3Measurement precision
If deep learning-based phase unwrapping is implemented, then accuracy is improved, but run-time efficiency decreases
Solution Approach 1:
The patent applies segmentation by dividing the 3D phase unwrapping task into separate 2D convolutional operations applied across time frames. This allows parallel processing of spatial dimensions while maintaining temporal correlations, significantly reducing computational complexity and run-time while preserving the accuracy benefits of deep learning-based phase unwrapping.
Data Source
AI summary
A method is disclosed for phase contrast magnetic resonance imaging (MRI) comprising: acquiring phase contrast 3D spatiotemporal MRI image data; inputing the 3D spatiotemporal MRI image data to a three-dimensional spatiotemporal convolutional neural network to produce a phase unwrapping estimate; generating from the phase unwrapping estimate an integer number of wraps per pixel; and combining the integer number of wraps per pixel with the phase contrast 3D spatiotemporal MRI image data to produce final output.