AI Fat Suppression in MRI Using Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional fat suppression techniques in MRI, such as fat saturation and Dixon methods, face challenges like non-uniformity due to B0 inhomogeneity and longer acquisition times, which affect image quality and resolution, especially in liver imaging.
Innovation Solution
An AI-based framework combining deep learning neural networks with multi-echo Dixon techniques is used to generate synthetic fat-suppressed images without the need for fat saturation during acquisition, utilizing water images as ground truth for training the network.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If Dixon techniques are used to separate fat and water signals, then uniformity of fat suppression is improved, but acquisition time increases
Solution Approach 1:
The patent applies preliminary action by performing fat-water separation using Dixon techniques during the acquisition phase, then pre-calculating and storing fat maps and water images before the actual imaging sequence. This allows the neural network to generate fat-suppressed images rapidly during reconstruction without requiring extended acquisition time, as the separation work was done in advance.
Solution Approach 2:
The patent uses copying by creating synthetic fat-suppressed images through a neural network that learns from training data containing actual fat-suppressed images. The network generates copies of fat-suppressed images from non-fat-suppressed input images, eliminating the need to acquire separate fat-suppressed images during the imaging session.
2Loss of time
If fat saturation is used to suppress fat signal, then acquisition time is reduced, but uniformity of fat suppression deteriorates due to B0 inhomogeneity
Solution Approach 1:
The patent introduces an intermediary approach by using a neural network as a mediator between non-fat-suppressed images and desired fat-suppressed images. Instead of directly applying fat saturation pulses that are sensitive to B0 inhomogeneity, the neural network learns the transformation from non-suppressed to suppressed images, providing uniform fat suppression without the limitations of conventional fat saturation methods.
Solution Approach 2:
The patent replaces the mechanical/physical fat saturation system (RF pulses and gradient fields) with an information-processing system (neural network). This substitution eliminates the physical limitations of B0 inhomogeneity sensitivity while maintaining rapid acquisition, as the neural network processes image data rather than relying on precise magnetic field manipulation.
3Measurement precision
If multi-echo Dixon acquisition is performed to correct B0 inhomogeneity and T2* effects, then measurement precision of fat fraction is improved, but productivity decreases
Solution Approach 1:
The patent applies partial action by using a simplified single-echo Dixon approach combined with neural network processing, rather than performing full multi-echo Dixon acquisition. The neural network compensates for the reduced echo information by learning from training data, achieving adequate fat fraction measurement precision with fewer echoes, thus improving imaging throughput.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach provides uniform fat suppression, reduces acquisition time, and maintains image quality by generating volumetric fat-suppressed images with a single echo time, while being insensitive to T1 effects and contrast-induced changes.
Implementation Method 1
Magnetic Resonance Imaging (MRI)
Implementation Method 2
separate fat and water signals
Data Source
AI summary
A computer-implemented method for using machine learning to suppress fat in acquired MR images includes receiving multi-echo images from an anatomical area of interest acquired using an MRI system. A first subset of the multi-echo images is acquired prior to application of contrast to the anatomical area of interest and a second subset of the multi-echo images is acquired after application of contrast to the anatomical area of interest. Next, data is generated including water images, fat images, and effective R*2 maps from the multi-echo images. The water images, the fat images, and the effective R*2 maps are used to create synthetic fat suppressed images. A neural network is trained to use the multi-echo images as input and the synthetic fat suppressed images as ground truth. A plurality of components of the neural network are saved to allow later deployment of the neural network on a computing system.