Poisson Resampling for Scintigraphic Image Resize
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for resizing nuclear medicine images fail to preserve the photon count statistics and noise characteristics, leading to suboptimal image quality in scintigraphic imaging.
Innovation Solution
The proposed method involves up-sampling and down-sampling nuclear medicine images using Poisson resampling to correct for excess photon counts and maintain realistic noise properties, while also employing machine learning for image count enhancement.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional interpolation methods (nearest neighbor, bilinear, bicubic) are used for up-sampling, then image resolution is improved, but photon count statistics and noise characteristics are distorted
Solution Approach 1:
The patent applies Poisson resampling to transform the statistical distribution of pixel values during up-sampling, changing the parameter distribution from artificial interpolation patterns to Poisson-distributed counts that reflect true photon counting statistics. This ensures that the resampled image maintains realistic noise characteristics while achieving higher resolution.
Solution Approach 2:
The patent replaces conventional mechanical interpolation algorithms with a stochastic Poisson resampling process. Instead of deterministic pixel value calculation based on neighboring pixels, the method uses probabilistic sampling from Poisson distributions to generate new pixel values, thereby substituting a physically unrealistic mechanism with one that accurately models photon detection statistics.
2Productivity
If down-sampling by averaging neighboring pixels is applied, then image processing speed is improved, but photon count conservation is compromised
Solution Approach 1:
The patent creates a copy of the high-resolution image at the target lower resolution by selectively sampling and summing photon counts from corresponding regions, rather than simply averaging pixel values. This copying process preserves the integer nature of photon counts and their Poisson statistical properties while achieving down-sampling.
Solution Approach 2:
The method changes the down-sampling operation from a deterministic averaging process to a probabilistic photon count summation process. By modeling the down-sampling as a photon collection process where counts are summed according to Poisson statistics, the patent preserves both the quantity of photon counts and their statistical characteristics.
3Reliability
If image acquisition time is increased to improve count statistics, then image quality is improved, but patient comfort and clinical throughput deteriorate
Solution Approach 1:
The patent performs Poisson resampling as a preliminary processing step to generate synthetic low-count images from high-count reference images. These pre-generated low-count images can then be used for training machine learning models or for comparison purposes without requiring actual long-duration acquisitions, thereby eliminating the time penalty while preserving statistical validity.
Solution Approach 2:
The method creates synthetic copies of low-count images through Poisson resampling of high-count reference images. These copied images replicate the statistical properties of true low-count acquisitions without requiring actual prolonged exposure times, thus providing a time-efficient alternative for generating statistically valid low-count images.
4Reliability
If machine learning enhancement is applied to increase effective sensitivity, then image quality is improved, but computational complexity increases
Solution Approach 1:
The patent performs Poisson resampling as a preliminary step before applying machine learning enhancement. By first generating images with realistic Poisson noise characteristics, the subsequent machine learning model can be trained more effectively on data that accurately represents the statistical properties of nuclear medicine images, thereby improving the overall effectiveness of the enhancement pipeline.
Solution Approach 2:
The method changes the input data characteristics for machine learning by applying Poisson resampling, which transforms the noise distribution to match true photon counting statistics. This parameter change in the data distribution enables the machine learning model to learn more accurate enhancement mappings, improving effective sensitivity despite the added computational step.
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 approach effectively preserves the photon count statistics and noise characteristics, improving the quality of scintigraphic images and enabling the generation of pseudo-planar images from SPECT projections, which can reduce image acquisition times and enhance patient experience.
Implementation Method 1
Poisson counting statistics play a visually perceivable and mathematically significant role in the image noise. As dictated by Poisson counting statistics, the variance is the signal, which is equal to the mean (expected true counts) of the sample
Data Source
AI summary
The present invention describes methods and techniques to improve scintigraphic images obtained by using nuclear medicine techniques for diagnostic analysis. The present invention describes a method and technique to up-sample and down-sample nuclear medicine images that models photon counts and noise characteristics of an image at its target resolution.