Trained Generator Function for Noise Reduction in Medical Imaging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In medical imaging, particularly in digital subtraction angiography, there is a challenge in achieving good image quality while minimizing x-ray dose, as high signal-to-noise ratios can lead to noise artifacts and require higher x-ray doses, and existing noise suppression methods can alter image impressions or introduce artifacts.
Innovation Solution
A computer-implemented method using a trained generator function based on a GA algorithm to create a differential image dataset from a real image dataset, reducing noise levels without the need for additional mask images, thereby allowing lower x-ray doses while maintaining image quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If the x-ray dose is reduced, then the radiation load is decreased, but the image quality deteriorates and the signal-to-noise ratio increases
Solution Approach 1:
The patent creates a synthetic mask image that copies the essential structural information from the real mask image but with suppressed noise. The generator function learns to reproduce the anatomical structures while filtering out the high-frequency noise, thereby maintaining image quality without requiring the actual high-dose mask image.
Solution Approach 2:
The patent introduces a trained generator function as an intermediary between the real mask image and the differential image generation process. This intermediary component processes the real mask image to produce a synthetic version with reduced noise, allowing the system to work with lower-dose images while maintaining quality through the mediating transformation.
2Measurement precision
If noise suppression methods are applied, then the signal-to-noise ratio is improved, but the image impression is altered and artifacts are introduced
Solution Approach 1:
The patent transforms the noise suppression problem from direct image processing to parameter space optimization. By training the generator function on pairs of real and synthetic images, the system learns optimal parameter transformations that suppress noise while preserving the essential visual characteristics and anatomical structures, avoiding the artifacts introduced by traditional filtering methods.
3Measurement precision
If a real image dataset with low noise level is used, then the noise level in the differential image dataset is reduced, but the x-ray dose must be increased
Solution Approach 1:
The patent creates a synthetic low-noise version of the real image dataset through the trained generator function. Instead of acquiring new low-noise images with higher x-ray dose, the system copies the essential information from the available real images while generating a cleaner version computationally, thereby achieving low noise levels without additional radiation exposure.
4Measurement precision
If optimized protocols are used, then the signal-to-noise ratio is improved, but the image pixel values differ across facilities and protocols
Solution Approach 1:
The patent develops a universal generator function that can process images from different x-ray facilities and protocols. By training on diverse datasets and learning the underlying anatomical structures rather than protocol-specific characteristics, the system achieves protocol-agnostic noise suppression that maintains consistency across different imaging conditions and facilities.
Data Source
AI summary
In an embodiment, a first real image dataset of an examination volume is received. The examination volume includes a vessel here, and the first real image dataset maps the examination volume includes contrast medium. Furthermore a differential image dataset of the examination volume is determined by application of a first trained generator function to input data. Here the input data includes the first real image dataset and a parameter of the trained generator function based on a GA algorithm. Furthermore the differential image dataset is provided.