Neural Network Fat Suppression in MR Imaging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for fat suppression in magnetic resonance (MR) imaging, such as spectral fat saturation and inversion pulses, either lead to artifacts or prolong imaging times, reducing the signal-to-noise ratio.
Innovation Solution
A method utilizing a trained artificial neural network, specifically a U-net architecture, to generate fat-reduced MR images by suppressing unwanted fat signal components while retaining the water signal, without significantly increasing acquisition time or reducing the signal-to-noise ratio.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-generated harmful factors
If spectral fat saturation methods are used, then the fat signal is suppressed, but artifacts occur in the MR images
Solution Approach 1:
The patent replaces conventional RF pulse-based fat saturation methods with a neural network-based post-processing approach. The neural network is trained to recognize and suppress fat signal components in MR images without introducing the artifacts that plague traditional spectral saturation methods, thereby improving image quality while maintaining fat suppression capability.
Solution Approach 2:
The patent introduces a neural network as an intermediary between the raw MR signal and the final image output. This neural network processes the MR images to selectively suppress fat signal components while preserving water signal integrity, acting as a mediator that eliminates the harmful fat signal without the artifacts generated by direct RF pulse methods.
2Object-generated harmful factors
If inversion pulse method is used, then fat components are reduced, but imaging time is prolonged
Solution Approach 1:
The patent applies preliminary action by training the neural network offline before actual imaging. The network learns fat suppression patterns from training data, enabling it to perform rapid fat suppression during actual imaging without requiring additional wait times or inversion pulses during the scanning process, thus maintaining fast imaging speeds.
Solution Approach 2:
The patent substitutes the time-consuming inversion pulse sequence with a computationally efficient neural network inference process. While the neural network requires training time, the actual imaging and fat suppression occur rapidly during the scanning process, eliminating the need for extended wait times associated with inversion recovery methods.
3Object-generated harmful factors
If inversion pulse method is used, then fat components are reduced, but signal-to-noise ratio is reduced
Solution Approach 1:
The patent applies local quality by enabling the neural network to selectively process different regions of the MR image with different operations. The network identifies and suppresses fat signal components in specific locations while preserving water signal regions, thereby maintaining high signal-to-noise ratio in the water signal while eliminating fat artifacts locally.
Solution Approach 2:
The neural network acts as an intermediary that selectively filters fat signal components without affecting the water signal intensity. Unlike inversion pulses that uniformly affect all signals, the neural network precisely targets fat signal regions, preserving the signal-to-noise ratio of the water signal while achieving fat suppression.
Data Source
AI summary
In a method for determining a fat-reduced MR image, a first MR image is provided having, apart from the other tissue constituents, MR signals from only one of the two fat constituents, the first MR image is applied to a trained ANN, which was trained by first MR training data as the input data, the training data including, apart from the other tissue constituents, MR signals from only the one of the two fat constituents, and using second MR training data as a base knowledge, the second MR training data including, apart from the other tissue constituents, no MR signals from the two fat constituents; and an MR output image is determined from the trained ANN, to which the first MR image was applied, as a fat-reduced MR image, wherein the fat-reduced MR image includes, apart from the other tissue constituents, no MR signals from the two fat constituents.