Medical Image Registration Using Variance-Based Regularization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing medical image processing techniques face challenges in accurately registering novel medical image datasets with reference datasets due to variations in patient physiology, leading to labor-intensive manual segmentation and subjective results, as well as increased computational intensity with multi-atlas approaches.
Innovation Solution
A computer system and method that determine a registration mapping between a novel medical image and a reference image, using variance data from a plurality of training images to apply regularization constraints, such as elastic and viscous fluid constraints, to restrict distortion and optimize registration, allowing for more accurate and efficient segmentation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual segmentation is performed by experts, then segmentation reliability is improved, but labor intensity and time consumption increase
Solution Approach 1:
The patent applies preliminary action by pre-segmenting a reference image dataset before registration. The reference dataset is segmented in advance using expert manual segmentation, and this pre-computed segmentation is then transferred to the novel dataset through registration. This eliminates the need for real-time expert segmentation during the registration process, maintaining reliability while improving productivity.
Solution Approach 2:
The patent uses copying by transferring segmentation information from the reference dataset to the novel dataset through spatial mapping. The segmentation labels and anatomical region definitions from the reference image are copied and applied to the novel image via the computed registration transformation, avoiding repeated manual segmentation while preserving expert-level accuracy.
2Productivity
If automated segmentation is performed using registration with a reference atlas, then productivity is improved, but manufacturing precision deteriorates due to physiological variations between patients
Solution Approach 1:
The patent applies local quality by allowing different degrees of freedom in the registration transformation for different spatial locations. The transformation model permits greater local deformation in regions where physiological variation is expected (such as abdominal organs) while maintaining stricter constraints in regions requiring higher precision. This location-dependent flexibility improves registration accuracy without sacrificing automation.
3Manufacturing precision
If multi-atlas approaches are used to handle physiological variations, then registration accuracy is improved, but computational intensity increases
Solution Approach 1:
The patent applies parameter changes by modifying the transformation model parameters to include location-dependent degrees of freedom. Instead of using multiple atlases with identical rigid transformation models, the system uses a single reference atlas with an enhanced transformation model that allows selective flexibility at different spatial locations. This reduces computational intensity while maintaining the ability to handle physiological variations.
4Device complexity
If rigid transformation models are used for registration, then device complexity is reduced, but adaptability deteriorates due to inability to accommodate anatomical variations
Solution Approach 1:
The patent applies dynamics by transitioning from a static rigid transformation model to a dynamic transformation model with location-dependent degrees of freedom. The transformation adapts its flexibility based on the spatial location being registered, allowing the model to accommodate anatomical variations in different body regions while maintaining computational tractability. This dynamic approach balances complexity and adaptability.
Data Source
AI summary
Certain embodiments provide a computer system for determining a registration mapping between a novel medical image and a reference medical image, the computer system comprising: a storage system adapted to store data representing the novel medical image and the reference medical image and variance data for a plurality of different locations in the reference medical image representing a statistical variation for corresponding locations identified in a plurality of training medical images; and a processor unit operable to execute machine readable instructions to determine a registration mapping between the novel medical image and the reference medical image in a manner that takes account of the variance data for the plurality of different locations in the reference medical image.