Automated PET-CT Registration via Neural Network Pseudo Imaging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The co-registration of PET and CT imaging data in cardiac imaging is a time-consuming and subjective process, prone to manual registration errors due to misregistration issues caused by patient repositioning and respiratory phase differences, even when acquired on a single machine.
Innovation Solution
A computer-implemented method using a neural network to generate pseudo imaging data associated with one modality from data of another modality, employing a conditional generative adversarial network (cGAN) and diffeomorphic registration algorithms for automated alignment of PET and CT imaging data, enabling efficient and accurate registration without requiring expert intervention.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual co-registration of PET and CT imaging data is performed, then registration accuracy can be achieved through expert operator expertise, but the process is time-consuming and subjective
Solution Approach 1:
The patent replaces the manual mechanical registration process with an automated computational system. A neural network automatically processes PET and CT imaging data, generating transformed PET images that are co-registered with CT images without requiring manual intervention. This substitution of manual operations with automated algorithms significantly reduces registration time while maintaining accuracy through learned transformation models.
Solution Approach 2:
The patent creates a transformed copy of the PET image data through the neural network. The neural network learns to transform PET images into a space that matches the CT image space, creating a virtual copy of the PET data that is automatically aligned with the CT anatomy. This copying approach allows accurate registration without moving or manipulating the actual imaging data physically.
2Measurement precision
If manual co-registration of PET and CT imaging data is performed, then detailed anatomical alignment can be achieved, but the process requires great operator expertise and is subjective
Solution Approach 1:
The patent replaces the complex manual registration process requiring expert knowledge with an automated neural network system. The neural network automatically learns the transformation rules for aligning PET and CT images, eliminating the need for operators to manually identify landmarks and adjust registrations. This automated approach removes subjectivity while maintaining the precision of anatomical alignment through data-driven learning.
3Measurement precision
If traditional manual registration techniques are used, then anatomical landmarks can be identified and aligned, but registration errors occur due to patient repositioning and respiratory phase differences
Solution Approach 1:
The patent replaces manual landmark-based registration with a neural network approach that processes the entire image volume. The neural network automatically identifies and aligns anatomical structures without relying on discrete landmarks that may be difficult to identify consistently. This holistic approach improves reliability by capturing the overall anatomical relationship between PET and CT images, compensating for patient repositioning and respiratory variations through learned transformation models.
Solution Approach 2:
The patent transforms the registration problem from manual landmark alignment to automated parameter learning. The neural network learns transformation parameters that map PET image space to CT image space, automatically adjusting for patient repositioning and respiratory phase differences. This parameter-based approach improves reliability by continuously adapting to variations in patient anatomy and positioning without manual intervention.
Data Source
AI summary
Automatic registration of multi-modal coronary imaging data is disclosed. First imaging data acquired using a first modality (e.g., positron emission tomography (PET) imaging) is applied to a neural network to output pseudo imaging data that is associated with a second modality (e.g., computed tomography (CT) imaging). The pseudo imaging data is then compared with (e.g., via nonlinear diffeomorphic registration) second imaging data acquired using the second modality to generate transformation information. This transformation information can then be applied to the first imaging data or other imaging data acquired in the first modality to register that imaging data with the second imaging data or other imaging data acquired in the second modality.