X-ray Noise Reduction via Parameterized Preprocessing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current noise reduction techniques for low-dose X-ray images, especially those based on deep learning, lack controllability and reliability in achieving desired noise reduction attributes, often resulting in artifacts and unpredictable outcomes.
Innovation Solution
A physics-based approach that utilizes correlated noise and noise adjustments through noise variance stabilization during training and application, allowing for controlled noise reduction by modifying noise attributes in preprocessing and postprocessing stages, ensuring predictable outcomes and improved image quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If deep learning-based noise reduction algorithms are used, then noise reduction performance is improved, but controllability and reliability deteriorate
Solution Approach 1:
The patent applies parameter changes by modifying the noise parameter (sigma) in the preprocessing stage of the deep learning algorithm. By adjusting this parameter, the noise reduction strength can be controlled in a predictable manner, transforming the uncontrollable deep learning process into one with adjustable and reliable parameters.
Solution Approach 2:
The patent introduces an intermediary mechanism through noise variance stabilization and parameter adjustment layers that mediate between the input image and the deep learning algorithm. This intermediary allows controlled manipulation of noise characteristics before processing, enabling reliable control over the noise reduction outcome.
2Manufacturing precision
If deep learning-based noise reduction algorithms are used, then noise reduction performance is improved, but predictability and comprehensibility deteriorate
Solution Approach 1:
By changing the noise parameter in preprocessing, the patent creates a comprehensible link between the input parameter and the noise reduction effect. This makes the otherwise black-box deep learning process understandable through its input parameters.
Solution Approach 2:
The patent performs preliminary action by stabilizing noise variance and adjusting parameters before the deep learning processing. This preprocessing step makes the subsequent complex processing predictable and comprehensible by establishing known initial conditions.
3Object-affected harmful factors
If lower X-ray dose is used, then radiation exposure is reduced, but noise increases and image quality deteriorates
Solution Approach 1:
The patent converts the harmful noise in low-dose images into a controllable parameter. By using noise variance stabilization and adjustable noise parameters in preprocessing, the noise that normally degrades image quality becomes a manageable aspect that can be optimized through parameter selection.
Solution Approach 2:
The patent applies parameter changes by adjusting the noise parameter in preprocessing to optimize the balance between noise reduction and image quality preservation in low-dose images, enabling effective noise management at lower radiation doses.
Data Source
AI summary
A method for noise reduction in a low-dose X-ray image includes preprocessing for determining input data, at least one trained function for determining noise-reduced output data from the input data, and postprocessing for determining a result image from the output data. At least one result parameter specifying at least one desired result attribute of the result image is received or determined. The at least one result attribute is obtained by modifying the preprocessing to set a noise value of at least one first noise parameter. The noise value is determined from the result parameter. The noise value may be selected to differ from a reference value of the first noise parameter. Alternatively or additionally, the at least one result attribute is obtained by setting, according to the result parameter, the at least one trained function to one of a plurality of predefined noise values of at least one second noise parameter.