Neural Network Compensation for Partial Fourier MRI Artifacts
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Partial Fourier techniques in magnetic resonance imaging often result in artifacts due to phase inconsistencies and limited acceleration of measurement data acquisition, restricting the use of higher PF factors and increasing measurement time.
Innovation Solution
A computer-implemented method using neural networks to generate compensation data, which simulates missing measurement data as if the entire k-space was sampled, thereby reducing artifacts and enhancing image sharpness without compromising image quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If traditional partial Fourier methods are used to reduce measurement time, then measurement time is reduced, but artifacts increase due to phase inconsistencies
Solution Approach 1:
A neural network is introduced as an intermediary between the undersampled k-space data and the final image reconstruction. The neural network processes the incomplete data, predicting and filling in missing information while maintaining phase consistency, thereby reducing artifacts that would normally result from the partial Fourier acquisition.
Solution Approach 2:
The patent changes the acquisition parameter by using a high partial Fourier factor (e.g., PF=0.75 or higher), which aggressively undersamples k-space to reduce measurement time. The neural network then compensates for this aggressive undersampling by learning to reconstruct the missing data patterns, allowing the system to operate at parameter extremes that would normally produce severe artifacts.
2Productivity
If higher PF factors are used to accelerate data acquisition, then measurement time is reduced, but image quality deteriorates due to limited acceleration
Solution Approach 1:
The neural network learns to copy or replicate the patterns of complete k-space data from training examples. By studying fully sampled data during training, the network learns the typical relationships between different k-space regions and can then generate accurate copies of missing data in the undersampled case, preserving image quality despite aggressive acceleration.
Solution Approach 2:
The neural network is trained in advance on a large dataset of complete k-space acquisitions. This preliminary training phase allows the network to learn the statistical patterns and physical constraints of MRI data before being deployed for actual accelerated scanning. The pre-learned knowledge enables the network to effectively reconstruct images from highly undersampled data without compromising quality.
3Manufacturing precision
If complete k-space sampling is performed to maintain image quality, then image sharpness is preserved, but measurement time increases
Solution Approach 1:
Instead of performing complete k-space sampling, the patent deliberately uses partial action by acquiring only a portion of the k-space data (e.g., 75% or more with PF factor ≥0.75). The neural network then performs the complementary action of reconstructing the missing portions, achieving the desired image sharpness without the time cost of complete sampling. This represents a strategic division of labor between physical acquisition and computational reconstruction.
Data Source
Figure 1~2
Figure 3~4
Figure 5
AI summary
A computer-implemented method according to the invention for creating adjustment data which compensates for missing measurement data in a measurement data set acquired using a partial Fourier technique (PF technique), i.e., measurement data that was not recorded although required for a complete measurement data set according to Nyquist, comprises the steps: - Receiving input data created on the basis of measurement data from the measurement data set, - Applying at least one trained adjustment function to the input data, whereby a comprehensive output data set is determined which compensates for measurement data missing in the measurement data set, - Providing the output data.