Deep Learning Multispectral MRI Enhancement Near Metal Implants
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Magnetic resonance imaging (MRI) in the presence of metallic implants is confounded by inhomogeneities in the polarizing magnetic field, leading to reduced effective image resolution and blurring due to voxel shifts in multispectral imaging, which are exacerbated by the need for further encoding to account for magnetic field inhomogeneities.
Innovation Solution
A neural network is trained on multispectral data from a metal-containing region and high-resolution MRI data from an overlapping artifact-free region to enhance spatial resolution and reduce noise, using techniques such as bin combination, super-resolution, and contrast transformation, replacing conventional algorithms with deep learning models like CNNs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multispectral imaging acquisitions are performed to account for magnetic field inhomogeneities caused by metallic implants, then image accuracy in metal-containing regions is improved, but spatial resolution deteriorates due to voxel shifts and blurring in the frequency encoding direction
Solution Approach 1:
The patent introduces a deep learning model as an intermediary between the low-resolution multispectral imaging data and the desired high-resolution output. The model is trained on pairs of low-resolution multispectral images and corresponding high-resolution reference images, learning to map between these representations and generate artifact-reduced, high-resolution images from multispectral acquisitions.
Solution Approach 2:
The patent changes the resolution parameter by applying super-resolution techniques through deep learning. The model learns to upscale low-resolution multispectral images to high resolution while simultaneously reducing metal-induced artifacts, effectively transforming the image quality parameters beyond what traditional reconstruction methods achieve.
2Reliability
If further encoding is applied to account for magnetic field inhomogeneities, then diagnostic accuracy in the presence of metal implants is improved, but acquisition duration increases
Solution Approach 1:
The patent performs preliminary training of the deep learning model using pairs of multispectral and reference images before actual diagnostic imaging. This pre-trained model can then rapidly process new multispectral acquisitions without requiring extended encoding sequences, reducing acquisition time while maintaining diagnostic accuracy through the learned artifact reduction capabilities.
Data Source
AI summary
Enhanced multispectral data, spectral bin images reconstructed therefrom, and/or composite images generated from spectral bin images are generated using deep learning-based techniques. As one example, a bin combination approach can be used to improve spatial resolution. As another example, a super-resolution technique can be used to improve spatial resolution. As yet another example, a contrast transformation technique can be used to generate images with a different contrast weighting network from multispectral data acquired from a metal-containing region.