Pseudo PET Image Generation from MRI Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current diagnostic methods using PET images are costly and involve radioactive substances, posing health risks and limitations in accessibility, especially for individuals with certain diseases like kidney disease, and are limited by the availability of scans from different modalities.
Innovation Solution
A system and method that uses a machine learning-based image processing model to generate pseudo PET images from MRI images, reducing the need for actual PET images and improving diagnostic accuracy by estimating PET image data from MRI data, thereby reducing costs and health risks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If PET imaging using radioactive medicines is performed to improve diagnostic accuracy, then diagnostic precision is improved, but financial burden and health risks increase
Solution Approach 1:
The patent generates pseudo-PET images as copies of actual PET images by training a machine learning model on paired PET-MRI data. The model learns the mapping relationship between MRI and PET images, then uses MRI images to generate synthetic PET images that replicate the diagnostic information without requiring actual radioactive tracer administration.
2Measurement precision
If PET imaging is performed to improve diagnostic accuracy, then diagnostic precision is improved, but cost increases
Solution Approach 1:
The patent replaces expensive PET imaging with a cost-effective machine learning model that runs on existing MRI data. The model, once trained, can generate pseudo-PET images at minimal computational cost, eliminating the need for expensive radioactive tracers and PET scanner usage fees while maintaining diagnostic accuracy.
3Measurement precision
If PET imaging is performed to improve diagnostic accuracy, then diagnostic precision is improved, but accessibility decreases due to contraindications
Solution Approach 1:
The patent introduces MRI as an intermediary modality that can be safely performed on patients with kidney disease, then uses the machine learning model to translate MRI images into pseudo-PET images. This intermediary approach allows diagnostic information to be obtained without directly administering radioactive substances to contraindicated patients.
4Measurement precision
If cross-modal synthesis is performed to improve diagnostic accuracy, then diagnostic precision is improved, but availability of corresponding scans is limited
Solution Approach 1:
The patent enables the MRI modality to serve itself by generating pseudo-PET images through the machine learning model. Instead of requiring separate PET scans, the system allows MRI images to self-generate the complementary diagnostic information they would otherwise need from PET imaging, making the process self-sufficient and eliminating dependency on PET scan availability.
Data Source
AI summary
A system (10) that provides diagnosis support information (110) relating to a disease of a target subject (5) includes: an acquisition unit (11) that acquires subject information (105) including actual image data (15) of an MR image including at least a reference region including part of an evaluation target region of the subject; and an information providing unit (12) that provides diagnosis support information (110) based on pseudo PET image data (115) of the evaluation target region generated by an image processing model (60) machine learned with training data (70) including actual image data (71) of a MR image of a reference region and actual image data (72) of a PET image including the evaluation target region of a plurality of test subjects so as to generate pseudo PET image data (75) of the evaluation target region from actual image data (71) of an MR image of the reference region, from the actual image data (15) of an individual MR image of the target subject.