Synthetic Radiologic Image Generation Without High-Dose Contrast Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for generating synthetic contrast-enhanced radiological images require target data from administering a standard amount of contrast agent, which is not available, limiting the ability to generate images with higher contrast.
Innovation Solution
A generative machine learning model is trained using reference representations of examination areas with and without contrast agent, allowing it to generate synthetic representations by subtracting or adding contrast, and adjusting parameters to minimize deviations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If a generative machine learning model is trained using reference representations with standard amount of contrast agent, then the model can generate synthetic representations, but the training cannot be carried out if target data with standard amount is not available
Solution Approach 1:
The patent creates synthetic copies of contrast-enhanced images by training a generative machine learning model on native images and using it to generate synthetic contrast-enhanced representations. This copying approach allows the model to learn the transformation from native to contrast-enhanced images without requiring actual standard-dose contrast agent administration, thereby resolving the data availability issue while maintaining training feasibility.
Solution Approach 2:
The patent changes the parameter of contrast agent amount by generating synthetic images with varying contrast enhancement levels. The model learns the relationship between native images and contrast-enhanced images with standard amounts, then can generate synthetic representations with different contrast levels, effectively decoupling the training requirement from the actual administration requirement.
2Illumination intensity
If a larger than standard amount of contrast agent is administered to generate target data, then synthetic representations with higher contrast can be generated, but the amount of contrast agent increases
Solution Approach 1:
The patent uses synthetic copying to create high-contrast images without actual high-dose contrast agent administration. The generative machine learning model learns the contrast enhancement pattern from standard-dose images and generates synthetic representations that simulate the appearance of high-contrast images, thereby achieving high contrast intensity while avoiding increased contrast agent quantity.
Solution Approach 2:
The patent replaces the expensive and potentially harmful real contrast agent with a computationally generated synthetic representation. Instead of administering additional contrast agent to achieve higher contrast, the system uses a trained model to generate synthetic images that mimic the effect of high-dose contrast agent, effectively substituting a physical substance with a digital copy.
3Productivity
If machine learning models are trained based on available training data, then synthetic images can be generated, but the training procedure cannot be carried out if target data is not available
Solution Approach 1:
The patent performs preliminary training of the generative machine learning model using native images and standard-dose contrast-enhanced images. The model learns the transformation relationship during this preliminary phase, enabling it to subsequently generate synthetic contrast-enhanced images without requiring target data to be available for each specific case. This preliminary action establishes the model's capability while ensuring training feasibility.
Solution Approach 2:
The patent trains the model to copy the transformation pattern from native images to contrast-enhanced images. By learning this copying relationship during training, the model can generate synthetic representations for new images without requiring corresponding target data to be available, thereby ensuring both training feasibility and subsequent image generation capability.
Data Source
AI summary
Systems, methods, and computer programs disclosed herein relate to generating synthetic representations, such as synthetic radiologic images.