PET Sinogram Noise Reduction via Iterative Gradient Smoothing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current PET reconstruction methods, such as delayed coincidence window estimation, face challenges in accurately processing non-uniformly sampled data, leading to increased variance and noise in sinogram data, which traditional smoothing techniques like Fourier analysis fail to address effectively, especially when dealing with irregularly sampled data and masks.
Innovation Solution
A method involving iterative second-order central differences and gradient calculations is applied to random event data from PET scanners, updating the data using an artificial time parameter to minimize the L-2 norm of sinogram image gradients, allowing for flexible application with any mask and effective noise reduction in both 2D and 4D data sets.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If delayed coincidence window estimation is used for random event data, then correlation between paired photons is removed, but variance increases in the prompt data
Solution Approach 1:
The patent applies preliminary smoothing action to the delayed coincidence data before subtracting it from the prompt data. By pre-smoothing the random estimate using the iterative second-order central differences method, the variance in the final subtracted data is reduced while maintaining accurate random event estimation.
2Object-affected harmful factors
If traditional Fourier smoothing is applied to uniformly sampled data, then high frequency noise is removed, but it fails for irregularly sampled PET sinogram data
Solution Approach 1:
The patent changes the sampling parameter by transforming irregularly sampled PET sinogram data into uniformly sampled interpolated sinogram data. This parameter transformation enables the application of standard Fourier smoothing techniques to irregularly sampled data by temporarily converting it to a uniform sampling regime, smoothing in the Fourier domain, then back-transforming to the original irregular sampling grid.
Solution Approach 2:
The patent introduces an intermediary uniformly sampled interpolated sinogram as a bridge between the original irregularly sampled data and the Fourier smoothing operation. This intermediary representation in uniform sampling space acts as a mediator that enables standard smoothing techniques to be applied, which would otherwise be incompatible with irregular sampling patterns.
3Object-affected harmful factors
If Fourier method is used for global signal representation, then smoothing can be applied, but it fails for images with irregular mask
Solution Approach 1:
The patent applies local quality by performing smoothing operations separately for different angular ranges and radial positions in the sinogram data. Instead of applying a single global Fourier transform, the method processes data in localized regions that respect the irregular mask boundaries, allowing the smoothing to adapt to local mask characteristics while maintaining overall image quality.
Data Source
AI summary
A method and apparatus for smoothing random event data obtained from a Positron Emission Tomography (PET) scanner. The method includes obtaining initial random event data u(s, φ, t=0)=u0(s, φ), corresponding to t=0, calculating second-order central differences uss, uφφ with respect to s, φ, calculating a gradient ut, using ut=2(uss+uφφ)−λ(u−u0), where λ is a constant parameter, and updating the random event data using u(s, φ, t2)=u(s, φ, t1)+Δt ut, where Δt=t2−t1, t1=0 in a first iteration, and Δt is greater than 0. The method repeats the steps of calculating the second-order central differences, calculating the gradient, and updating the random event data until a change in u(s, φ, t) from a previous iteration is less than a predetermined threshold value.