Deformable Liver Image Registration via Boundary Alignment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Surface-based liver registration methods distort partial liver images and fail to optimally register tumors due to stretching and neglect of internal liver volumes, leading to inaccurate tumor tracking.
Innovation Solution
A deformable registration method that aligns common liver regions and landmarks, ignoring missing regions by using a deep learning model trained on liver landmarks and surfaces, which detects missing liver areas and focuses registration on shared regions, employing cropping, padding, and a registration model to compute deformation fields.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If surface-based liver registration is used, then the liver surface can be registered, but the internal liver volume is distorted and tumors are not registered optimally
Solution Approach 1:
The patent segments the liver registration problem into two distinct components: surface-based registration and volume-based registration. By separating these functions, the system can optimize each independently - using surface matching for boundary alignment while using volume-based deformable registration for internal structures and tumors, thereby resolving the contradiction between surface registration accuracy and internal volume registration accuracy
Solution Approach 2:
The patent introduces a deformable registration model as an intermediary between the surface registration and the internal volume/tumor registration. This intermediary layer computes deformation fields that map the source liver volume to the target liver volume, allowing tumors and internal structures to be accurately registered even when surface-based methods fail, thus mediating between surface accuracy requirements and internal volume accuracy requirements
2Productivity
If surface-based liver registration is used, then registration can be performed quickly, but partial liver images are stretched and distorted
Solution Approach 1:
The patent employs dynamic deformable registration that adapts to the specific characteristics of each liver volume. Rather than applying a fixed surface-based transformation, the system computes dynamic deformation fields that flexibly adjust to match internal liver structures and tumors, preserving the natural shape of partial liver images while maintaining registration speed through optimized computational approaches
3Manufacturing precision
If complete liver volume registration is attempted, then all liver regions can be aligned, but missing liver regions cause incorrect registration
Solution Approach 1:
The patent extracts and identifies missing liver regions from the source image, then excludes these regions from the registration computation. By removing problematic areas where data is incomplete or unreliable, the system performs registration only on reliably visible regions, preventing the propagation of errors that would otherwise occur when attempting to register missing or partially visible liver areas
Solution Approach 2:
The patent applies different registration strategies to different regions of the liver based on data quality. In regions where complete liver volume data is available, full deformable registration is applied. In regions where data is missing or incomplete, the system adapts by focusing only on reliably visible structures and landmarks, thereby maintaining high registration reliability across the entire liver while accounting for local data quality variations
Data Source
AI summary
Systems and computer-implemented methods of performing image registration. One method includes receiving a first image and a second image acquired from a patient at different times and, in each of the first image and the second image, detecting an upper boundary of an imaged object in an image coordinate system and detecting a lower boundary of the imaged object in the image coordinate system. The method further includes, based on the upper boundary and the lower boundary of each of the first image and the second image, cropping and padding at least one of the first image and the second image to create an aligned first image and an aligned second image and executing a registration model on the aligned first image and the aligned second image to compute a deformation field between the aligned first image and the aligned second image.