CNN Correction of Chemical Shift Artifacts in Bipolar Dixon MRI

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

The Dixon magnetic resonance imaging method using bipolar readout gradients suffers from chemical shift artifacts (CSA), which degrade image quality and require longer acquisition times, making it impractical for certain applications.

Innovation Solution

A deep learning-based correction method using convolutional neural networks (CNNs) is employed to remove CSA from MR data, allowing for the conversion of bipolar Dixon-MR acquisition data to match monopolar data quality without increasing acquisition time.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of time

If bipolar readout gradients are used in Dixon-MR imaging, then acquisition time is reduced, but chemical shift artifacts increase

Engineering Contradiction:

Improveacquisition timeVSAvoidchemical shift artifacts

Core Design Contradiction:

Loss of timeVSObject-affected harmful factors

Solution Approach 1:

A convolutional neural network (CNN) is introduced as an intermediary processing step between the bipolar Dixon-MR data acquisition and the final image reconstruction. The CNN takes the bipolar acquisition data as input and outputs corrected data with chemical shift artifacts removed, thereby enabling fast acquisition without sacrificing image quality.

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent replaces traditional artifact correction methods (which would require slower acquisition protocols) with a deep learning-based computational approach. The CNN model substitutes for mechanical changes in the acquisition process, allowing bipolar gradients to be used while correcting artifacts in the digital domain.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Manufacturing precision

If chemical shift artifacts are corrected using traditional methods, then image quality improves, but acquisition time increases

Engineering Contradiction:

Improveimage qualityVSAvoidacquisition time

Core Design Contradiction:

Manufacturing precisionVSLoss of time

Solution Approach 1:

The convolutional neural network is pre-trained on simulated bipolar Dixon-MR data with known ground truth images. This preliminary training allows the model to learn the mapping from artifact-contaminated bipolar data to clean images, enabling rapid correction during actual scanning without requiring time-consuming acquisition protocols.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent creates synthetic training data by copying and modifying bipolar acquisition patterns through computational simulations. These synthetic images are used to train the CNN, allowing the model to generalize and correct artifacts in real patient scans without requiring additional time for acquisition.

Inventive Principle:

Principle #26Copying

Data Source

PatentEP4113150B1Correction of chemical shift artifacts from bipolar dixon-mr imaging data using a convolutional neural network (CNN)

Publication Date: 2025.06.04 SIEMENS HEALTHINEERS AG

Figure 1
Figure 2
Figure 3

AI summary

The present invention relates to a method, a processing unit for a magnetic resonance imaging device for carrying out the method, and an associated computer program. The method serves to correct chemical displacement artifacts (CSA) that occur in magnetic resonance Dixon MR when using bipolar readout gradients (fast Dixon MR) to acquire in-phase and out-of-phase echoes. The method employs a trained CNN that has been trained with in-phase and out-of-phase acquisition data using the Dixon MR method. This data includes acquisition data containing CSA in opposite directions and acquisition data containing CSA in only one direction. The CNN is trained to transform the fast Dixon MR acquisition data so that it exhibits only CSA occurring in the same direction.