QSM Map Reconstruction Across Resolutions With Dipole Compensation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing QSM map reconstruction methods fail when input MRI images have resolutions different from the resolution for which the inference network was trained, leading to inaccurate reconstructions.
Innovation Solution
A method involving the generation of sub-images at a training resolution, assembly of QSM sub-maps, and application of dipole compensation to generate a QSM map with a desired resolution, using a neural network trained at a specific resolution.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If an inference network is trained at a specific resolution, then it can successfully reconstruct QSM maps for input images with that resolution, but it fails to reconstruct QSM maps for input images with different resolutions
Solution Approach 1:
The input image is divided into multiple sub-images that can be processed by the inference network trained at a specific resolution. Each sub-image is processed independently to generate corresponding QSM sub-maps, which are then assembled to form the complete QSM map at the original resolution.
Solution Approach 2:
The method changes the resolution parameter by resampling the input image to match the training resolution, allowing the fixed-resolution inference network to process images of various original resolutions. The dipole compensation further adjusts parameters to correct resolution mismatches.
2Manufacturing precision
If the inference network processes images at higher resolution, then the QSM map quality improves, but the computational complexity and processing time increase
Solution Approach 1:
The high-resolution input image is segmented into multiple lower-resolution sub-images that can be processed by the inference network. This reduces the computational load on each processing pass while maintaining the ability to reconstruct the full high-resolution QSM map through assembly of sub-maps.
Solution Approach 2:
The method processes the image in partial segments (sub-images) rather than attempting to process the entire high-resolution image at once. This allows the inference network to operate at its optimal resolution while still achieving high-quality reconstruction through multiple passes and dipole compensation.
3Productivity
If the inference network processes images at lower resolution, then the processing speed increases, but the QSM map reconstruction accuracy decreases
Solution Approach 1:
The method segments the processing into multiple stages: resampling to training resolution, processing by the inference network, assembly of sub-maps, and dipole compensation. This segmentation allows efficient processing at lower resolution while recovering accuracy through the compensation step.
Solution Approach 2:
The dipole compensation is applied as a preliminary correction step after assembly but before final output. This preliminary action corrects the resolution-related inaccuracies introduced during downsampling, ensuring accurate QSM map reconstruction even when processing at lower resolutions.
Data Source
AI summary
A method of generating a quantitative susceptibility mapping (QSM) map capable of reconstructing data having various resolutions using a given network trained at a specific resolution includes preparing a local field map having input resolution resolinput, generating a plurality of sub-images having training resolution resoltrain by resampling the local field map, inputting each of the sub-images into QSMnet_resoltrain, which is an inference network trained with the training resolution, acquiring a plurality of QSM sub-images having the training resolution resoltrain from the inference network, generating one QSM assembled map by assembling the plurality of QSM sub-images, and generating a QSM map (corrected QSM map) from the QSM assembled map by applying a dipole compensation method.