Synthetic Electron Density Imaging from MRI Using Transfer Coefficients
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing MRI-only workflows for radiotherapy planning face challenges in generating synthetic electron density images due to the lack of correspondence between MR image pixel intensities and tissue radiation attenuation properties, leading to ambiguity and the need for time-consuming and costly combined CT/MRI workflows.
Innovation Solution
A machine-learning model is trained to parameterize an image transfer function using MR images, generating synthetic electron density images without direct pixel mapping, allowing for automated and accurate generation of sCT images without the need for preparatory segmentation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If MRI-only workflow is used to generate synthetic electron density images, then the need for additional CT scans is eliminated (reducing radiation dose and cost), but the accuracy and reliability of electron density information deteriorates due to lack of correspondence between MR image intensities and radiation attenuation properties
Solution Approach 1:
The patent introduces an intermediate atlas-based mapping approach where pre-matched MR and CT atlases serve as a mediator to transfer electron density information. The deformation field from registering the patient's MR image to the atlas provides the transformation needed to map atlas-derived electron density values to the patient's anatomy, bridging the gap between MR intensity and electron density
Solution Approach 2:
The patent performs preliminary actions by pre-matching multiple MR and CT atlases before actual patient imaging. The atlases are pre-registered and pre-mapped to establish reliable correspondence relationships in advance, so that when a patient's MR image is acquired, the electron density information can be quickly and accurately transferred without requiring a CT scan at the time of treatment planning
2Object-affected harmful factors
If atlas-based approach is used to generate sCT images, then electron density information can be obtained without CT scans, but the processing time increases significantly
Solution Approach 1:
The patent performs preliminary actions by pre-matching multiple MR and CT atlases before actual patient imaging. The atlases are pre-registered and pre-mapped to establish reliable correspondence relationships in advance, so that when a patient's MR image is acquired, the electron density information can be quickly and accurately transferred without requiring a CT scan at the time of treatment planning
Solution Approach 2:
The system uses the patient's own MR image to generate the deformation field that drives the atlas-based mapping. The MR image serves multiple purposes: as the input for deformation field calculation and as the target for transferring electron density information, eliminating the need for separate processing steps
3Object-affected harmful factors
If segmentation-based approach is used to convert MR images to sCT, then electron density information can be generated, but the complexity of manual segmentation and tissue classification increases
Solution Approach 1:
The patent extracts the complex segmentation and classification tasks from the online processing workflow and relocates them to the offline atlas creation phase. The atlases are pre-segmented and pre-classified with ground truth electron density information, so that during patient-specific processing, only simple deformation field calculation and value transfer are needed, eliminating the need for manual segmentation at the time of treatment planning
Data Source
AI summary
A conversion device is operable to perform a learning-based method of generating a synthetic electron density image (sCT) of an anatomical portion based on one or more magnetic resonance (MR) images. The method is processing-efficient and capable of producing highly accurate sCT images irrespective of misalignment in the underlying training set. The conversion device receives and installs a machine-learning model trained to predict coefficients of an image transfer function. The conversion device then receives a current set of MR images of the anatomical portion, computes current coefficients of the image transfer function by operating the machine-learning model on the current set of MR images, and computes a current sCT image of the anatomical portion by operating the current coefficients, in accordance with the image transfer function, on the current set of MR images.