Deep Learning Landmark Alignment for MRI Stitching Artifacts
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Magnetic resonance imaging (MRI) systems face challenges in generating whole-body images due to limited scan range, leading to stitching artifacts like discontinuity, which can result in misdiagnosis, especially when the spine appears broken or misaligned at stitching locations between sectional scans.
Innovation Solution
A method involving multiple sectional scans is performed, with a trained deep learning network determining the location of landmarks in each scan, allowing for precise alignment and stitching of datasets based on localized information in the overlapped regions, thereby avoiding discontinuity in the constructed image.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If multiple sectional scans are performed to cover a large range of the image subject, then the coverage area is improved, but stitching artifacts like discontinuity occur at the boundaries between scans
Solution Approach 1:
The patent divides the whole-body imaging process into multiple sectional scans, each covering a specific anatomical section. By segmenting the imaging task into overlapping sections and then stitching them together using deep learning-based landmark alignment, the system achieves both large coverage area and maintains image continuity at boundaries.
2Productivity
If traditional stitching methods are used to combine sectional datasets, then the processing speed is improved, but stitching errors cause landmark discontinuity and potential misdiagnosis
Solution Approach 1:
The patent replaces traditional mechanical or algorithmic stitching methods with a deep learning-based approach. The trained neural network automatically identifies and aligns landmarks across sectional datasets, substituting manual or rule-based stitching mechanisms with an intelligent system that achieves both speed and precision through learned patterns.
3Productivity
If the scan range is increased to cover the whole body in a single scan, then the number of scans is reduced, but the MRI system's limited scan range prevents achieving sufficient resolution and accuracy
Solution Approach 1:
Rather than attempting a single whole-body scan that would compromise resolution, the patent segments the body into multiple anatomical sections scanned with high resolution. The segmentation approach allows each section to be imaged at full resolution while the deep learning stitching process combines them seamlessly, achieving both precision and whole-body coverage.
Data Source
AI summary
Various methods and systems are provided for scanning an image subject along a scan axis. The method includes stitching sectional datasets acquired from different anatomical sections of the image subject based on locations of a landmark in the sectional datasets.