Photon-Counting CT Noise Parameter Extraction for Compression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Photon-counting CT imaging systems generate large amounts of data, particularly due to the noisy channels, which limits the effective compressibility of the data, leading to higher storage and transmission demands.
Innovation Solution
An image processing method that replaces the noise channel with two noise parameter images, which are derived from denoised CT image data using normal distribution assumptions or machine learning algorithms, allowing for efficient compression and subsequent noise generation in the image domain.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the noise channel is stored in the compressed data, then the diagnostic image quality is maintained, but the compression rate is limited and storage requirements increase
Solution Approach 1:
The noise channel is extracted and removed from the data set before compression. Instead of storing the actual noise channel, the invention stores only the denoised image data and generates noise synthetically during display, thereby reducing storage requirements while maintaining the ability to produce diagnostic-quality images
Solution Approach 2:
The invention changes the representation of noise from storing actual noise values to storing parameters that define noise characteristics (mean and standard deviation). This parameter-based approach allows for efficient compression while enabling accurate noise reconstruction during image display
2Measurement precision
If the noise channel is compressed losslessly, then the image quality is preserved, but the compression factor is limited to 2 or lower
Solution Approach 1:
The noise channel is extracted and removed from the data set before compression. Instead of storing the actual noise channel, the invention stores only the denoised image data and generates noise synthetically during display, thereby reducing storage requirements while maintaining the ability to produce diagnostic-quality images
Solution Approach 2:
The invention changes the representation of noise from storing actual noise values to storing parameters that define noise characteristics (mean and standard deviation). This parameter-based approach allows for efficient compression while enabling accurate noise reconstruction during image display
3Productivity
If the noise channel is discarded completely, then the compression rate increases, but the image loses realistic appearance and diagnostic quality
Solution Approach 1:
The invention changes the representation of noise from storing actual noise values to storing parameters that define noise characteristics (mean and standard deviation). This parameter-based approach allows for efficient compression while enabling accurate noise reconstruction during image display
Solution Approach 2:
The noise characteristics (mean and standard deviation) are pre-calculated and stored in the compressed data. This preliminary preparation enables the synthetic noise generation to be performed efficiently during image display without compromising image quality or requiring additional processing time
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Image processing methods are provided for processing photon-counting CT images. The processing involves receiving denoised CT image data and, for each image voxel, deriving normal distribution mean and deviation values. Noise parameter images are generated from these per-voxel mean and deviation values and the denoised CT image data is compressed. During decompression, per-voxel mean and deviation noise values are obtained so that per-voxel noise is generated in the image domain. The per-voxel noise is applied to the decompressed denoised CT image.