Tumor-Aware Image Registration With Adaptive Volume Preservation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current deformable image registration methods fail to preserve tumor volumes accurately during alignment of medical images, which is crucial for tracking tumor growth and planning cancer treatment, as they prioritize image similarity over tumor properties.
Innovation Solution
A two-stage framework for deformable image registration that includes unsupervised tumor mask estimation and volume-preserving registration, using a similarity-based network to identify tumor regions and adaptively apply volume-preserving losses to maintain tumor volumes while maximizing image similarity in non-tumor areas.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional deformable registration methods are used to align medical images, then image similarity is improved, but tumor volume preservation deteriorates
Solution Approach 1:
The patent segments the image into tumor regions and non-tumor regions using a trained segmentation model. This allows differential treatment of different regions during registration, enabling image similarity optimization in non-tumor areas while preserving tumor volume through separate constraints.
Solution Approach 2:
The patent applies different quality requirements to different regions: image similarity is prioritized in non-tumor regions while volume preservation is prioritized in tumor regions. This is achieved through region-specific loss functions and constraints in the registration objective.
2Manufacturing precision
If learning-based registration methods are used to maximize image similarity, then alignment accuracy is improved, but tumor property preservation deteriorates
Solution Approach 1:
The patent implements local quality by applying different optimization criteria to different regions. Non-tumor regions are optimized for alignment accuracy through similarity metrics, while tumor regions are constrained to preserve volume and key properties, achieving both goals simultaneously through region-specific loss weighting.
Solution Approach 2:
The patent uses a two-stage approach where segmentation feedback from the first stage informs the registration process in the second stage. The segmentation model provides tumor region masks that guide the registration objective to differently treat tumor and non-tumor areas, creating a feedback loop that preserves tumor properties while maintaining alignment accuracy.
3Measurement precision
If regular registration is applied to align anatomy across different time periods, then anatomical alignment is improved, but tumor size tracking deteriorates
Solution Approach 1:
The patent segments tumor regions from anatomical structures using a trained segmentation network. This separation allows the registration to align anatomical structures for accurate anatomical alignment while independently preserving tumor volume for accurate tumor size tracking across time periods.
Solution Approach 2:
The patent applies different quality metrics to different structures: anatomical alignment is optimized in non-tumor regions while tumor volume is preserved in tumor regions. This enables simultaneous achievement of anatomical alignment and tumor size tracking through spatially varying optimization criteria.
Data Source
AI summary
Similarity-based deformable registration of medical images that contain tumor regions often undesirably leads to disproportionate volume changes to tumor regions. An image-registration framework for preserving tumor volumes while promoting similarity in non-tumor part is provided. The framework involves a two-stage process. In the first stage, similarity-based registration is used to identify potential tumor regions by their volume change, generating a soft tumor mask accordingly. In the second stage, a volume-preserving registration network is used. The network uses a novel adaptive volume-preserving loss that penalizes the change in size adaptively based on the masks calculated from the previous stage. The framework balances image similarity and volume preservation in different regions, i.e. non-tumor and tumor regions, by using soft tumor masks to adjust the imposition of volume-preserving loss on each one.