Medical Imaging Data Normalization for Sinogram Noise Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Medical imaging technologies, such as CT scans, face challenges with noise in the sinogram domain leading to artifacts like streaks and heavy-tailed noise in reconstructed images, particularly when dealing with highly attenuating objects, which are difficult to correct using image domain techniques alone.
Innovation Solution
A multi-stage approach involving normalization, deep learning-based denoising, and de-normalization of projection data to reduce noise and artifacts, utilizing a deep learning network to process data from medical imaging systems, including CT imaging, to generate artifact-free images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If image domain techniques are applied to reduce noise, then image quality may be improved, but noise in the sinogram domain leading to streaks and heavy-tailed artifacts cannot be effectively corrected
Solution Approach 1:
Instead of applying denoising techniques in the image domain after reconstruction, the patent inverts the approach by applying denoising in the sinogram domain before reconstruction. This allows direct correction of noise that causes streaks and heavy-tailed artifacts, addressing the root cause rather than treating symptoms in the final image.
Solution Approach 2:
The patent introduces an intermediate processing step in the sinogram domain between data acquisition and image reconstruction. This intermediary denoising operation targets the specific noise characteristics in the sinogram that lead to reconstruction artifacts, serving as a mediator between raw data and final image production.
2Measurement precision
If deep learning-based denoising is applied directly to projection data with huge dynamic range, then denoising performance may improve, but the huge dynamic range makes effective denoising difficult
Solution Approach 1:
The patent applies preliminary normalization to the projection data before feeding it to the deep learning denoising network. This preprocessing step transforms the data with huge dynamic range into a normalized form with reduced dynamic range, making the subsequent denoising operation more effective and stable.
Solution Approach 2:
The patent changes the parameter scale of the projection data through normalization, transforming values from a huge dynamic range into a compressed, normalized range. This parameter transformation enables deep learning models to process the data more effectively without being overwhelmed by the extreme variations in magnitude.
3Measurement precision
If normalization is applied to reduce dynamic range, then denoising effectiveness improves, but the original data scale and information may be altered
Solution Approach 1:
The patent performs preliminary normalization to enable effective denoising, then applies a reverse normalization step after denoising to restore the data to its original scale. This two-step process allows temporary transformation for processing while preserving the original data characteristics and information content.
Solution Approach 2:
The patent implements a feedback mechanism where normalized data is processed through denoising, then the result is fed back through reverse normalization to produce the final output. This feedback loop ensures that the benefits of normalization-based denoising are realized while maintaining fidelity to the original data scale and information.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This method effectively reduces noise and artifacts in medical imaging, improving image quality by addressing the challenges of high dynamic ranges and non-positive measurements, resulting in clearer and more accurate reconstructed images.
Implementation Method 1
de-noising the normalized data utilizing a deep learning-based denoising network
Implementation Method 2
the attenuated radiation impacts a detector where the attenuated intensity data is collected
Implementation Method 3
a detector produces signals representative of the amount or intensity of radiation impacting discrete pixel regions
Data Source
AI summary
A computer-implemented method for image processing is provided. The method includes obtaining data acquired by a medical imaging system. The method also includes normalizing the data. The method further includes de-noising the normalized data utilizing a deep learning-based denoising network. The method even further includes de-normalizing the de-noised data. The method yet further includes generating blended data based on both the data and the de-normalized de-noised data.