Simulated CT Image Generation via Tissue Class Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for generating simulated CT images from MR images in radiation therapy face challenges in accuracy and workflow efficiency, particularly in highly variable anatomical regions, as they often rely on simplistic tissue classification and ignore local density variations.
Innovation Solution
A method that retrieves MR image data, analyzes tissue types, registers specific reference data sets with Hounsfield Unit values, and computes simulated CT images by assigning these values to corresponding tissue types, accounting for location-specific density variations and anatomical variability through rigid and non-rigid registration.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If a CT-based density atlas is registered to the MR image to generate simulated CT images, then the workflow is simplified and additional CT radiation is avoided, but the accuracy deteriorates in highly variable anatomical regions such as the pelvic region due to inability to cover anatomical variations between patients
Solution Approach 1:
The method segments the MR image into multiple tissue classes (e.g., fat, muscle, bone, air) and performs separate registration of CT-based density atlases for each tissue class. This segmentation allows the system to handle different tissue types with their own specific density characteristics, improving accuracy in variable anatomical regions while maintaining workflow simplicity through automated processing.
Solution Approach 2:
The invention applies local quality by using tissue-class-specific density atlases that are registered separately for different anatomical regions and tissue types. Instead of using a single global atlas, the system tailors the density assignment to local tissue characteristics, thereby improving accuracy in highly variable regions like the pelvis while keeping the overall workflow efficient.
2Object-affected harmful factors
If MR images are used alone for radiation therapy planning without generating simulated CT images, then the radiation exposure to the patient is reduced, but the dosimetry accuracy deteriorates because MR intensities do not uniquely correspond to electron densities or attenuation coefficients
Solution Approach 1:
The invention introduces simulated CT images as an intermediary that bridges MR imaging and radiation dosimetry. The simulated CT images are generated by registering tissue-class-specific CT density atlases to the MR image, providing the necessary electron density information for accurate dosimetry while avoiding the need for additional CT scans, thus reducing patient radiation exposure.
Solution Approach 2:
The method changes the parameter representation by transforming MR intensity values into Hounsfield Units through atlas-based registration. This parameter transformation allows the system to use MR images for anatomical visualization while deriving accurate electron density information from the registered CT atlases, enabling precise dosimetry without additional CT radiation.
3Loss of time
If a single CT-based density atlas is registered to the entire MR image, then the processing time is reduced and the method remains computationally efficient, but the manufacturing precision deteriorates due to inability to account for tissue-specific density variations
Solution Approach 1:
The invention segments the density atlas registration process into multiple tissue-class-specific registrations. By dividing the single atlas registration into separate registrations for different tissue types (fat, muscle, bone, etc.), the system achieves higher density map accuracy while maintaining computational efficiency through optimized processing of each tissue class separately.
Solution Approach 2:
The method applies partial action by performing registration only for the specific tissue classes present in each region of interest, rather than uniformly processing the entire image with a single atlas. This selective approach reduces unnecessary computational overhead while improving accuracy where it matters most, balancing processing time and precision.
Data Source
AI summary
A device, system and method for generating one or more simulated CT images from MR images, including retrieving MR image data for one or more body parts of a living being, said MR image data including a plurality of pixels and/or voxels, analyzing said MR image data to identify one or more tissue and/or material types for one or more of said plurality of pixels and/or voxels, registering one or more reference data sets to said identified one or more tissue and/or material types, said reference data sets corresponding to a specific one of said identified tissue and/or material types, said reference data sets including reference values, and computing one or more simulated CT images by assigning said reference values to said pixels and/or voxels corresponding to said identified one or more tissue and/or material types.