Frequency Domain Weighting for Variable Contrast Radiological Imaging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for generating radiological images with variable contrast enhancement require extensive training data and are limited by the need for additional training when adjusting contrast levels, making them impractical for widespread medical use and prone to false positives/negatives due to statistical model generalizability issues.
Innovation Solution
A computer-implemented method that generates contrast-enhanced radiological images by transforming representations of an examination area in frequency space, applying a frequency-dependent weight function to amplify signal differences caused by varying contrast agent amounts, and combining these transformations to produce enhanced images in spatial space, allowing for adjustable contrast without extensive retraining.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If an artificial neural network is trained to predict radiological images with variable contrast enhancement, then the ability to generate images with different contrast levels is improved, but the requirement for extensive training data for each contrast level and additional training time increases
Solution Approach 1:
The patent creates synthetic training data by copying and transforming existing low-contrast images through frequency domain processing. A frequency-dependent weighting function is applied to amplify signal differences, generating artificial high-contrast images without requiring actual high-contrast scans. This copying approach allows the neural network to learn variable contrast enhancement from a single training set, eliminating the need for separate training data for each contrast level.
Solution Approach 2:
The patent changes the frequency domain parameters of existing images by applying a frequency-dependent weighting function. This transforms the spectral content to simulate different contrast enhancement levels. By manipulating frequency parameters rather than requiring different physical contrast agent doses, the system generates diverse training examples from a single dataset, resolving the contradiction between adaptability and training time.
2Measurement precision
If radiological images are enhanced with higher contrast, then the contrast between areas with and without contrast agents is improved, but the risk of false positives and false negatives increases due to statistical model generalizability limits
Solution Approach 1:
The patent introduces a frequency-dependent weighting function as an intermediary between the original image data and the enhanced output. This intermediary operates in the frequency domain to selectively amplify signal differences while maintaining a deterministic, traceable transformation process. Unlike direct statistical learning approaches, this intermediary ensures consistent and reliable contrast enhancement without introducing the generalizability errors associated with training on limited high-contrast data.
3Reliability
If a deterministic process is used for generating variable contrast enhancement, then the reliability and traceability are improved, but the ability to adapt to different contrast requirements may be limited
Solution Approach 1:
The patent implements a dynamic weighting function in the frequency domain that can be adjusted to produce different contrast enhancement levels. The frequency-dependent weights allow the deterministic process to adapt to various contrast requirements by modifying the transformation parameters. This dynamic approach maintains traceability through a consistent mathematical framework while providing the flexibility to generate images with different contrast characteristics from the same training set.
Data Source
Figure 1
Figure 2
Figure 3(a)~3(h)
AI summary
The present disclosure relates to the technical field of generating artificial contrast-enhanced radiological images.