Scatter Correction in PET Imaging Using Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for generating images from measurement data in oncologic imaging using ionizing radiation are prone to errors due to photon scattering, particularly in PET imaging, where scattered coincidences can distort the spatial distribution of emitted photons, leading to inaccurate image representation of radioactive imaging agents within the body.
Innovation Solution
A computer-implemented method that uses multiple sets of measured data with different energy bins and a deep convolutional neural network to correct for scatter radiation by iteratively adjusting preliminary images to a template image, taking into account the neighborhood of each detector element, to produce a more accurate representation of the radioactive imaging agent distribution.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional image generation methods are used without scatter correction, then the processing is simpler and faster, but the image accuracy deteriorates due to photon scattering distortions
Solution Approach 1:
The patent applies preliminary scatter correction by generating a scatter estimate before the main image reconstruction process. This scatter estimate is created using a simplified model that predicts scatter distribution based on the object's attenuation properties, allowing the main reconstruction algorithm to work with corrected data that has scatter effects pre-compensated for.
Solution Approach 2:
The patent introduces an intermediary scatter estimation step that acts as a mediator between the raw measured data and the final image reconstruction. This intermediary process creates a separate scatter distribution map that is then subtracted from the measured data, effectively decoupling the complex scatter physics from the main reconstruction algorithm.
2Measurement precision
If scatter correction is applied to improve image accuracy, then the measurement precision improves, but the processing time increases
Solution Approach 1:
The patent segments the image reconstruction process into distinct phases: a preliminary scatter estimation phase using a simplified model, and a main reconstruction phase using the corrected data. This segmentation allows computationally intensive scatter correction to be performed once beforehand, rather than iteratively during reconstruction, significantly reducing total processing time.
Solution Approach 2:
The patent changes the parameters used in scatter estimation by using simplified physical models with fewer variables during the preliminary phase, rather than full Monte Carlo simulations. This parameter simplification maintains adequate scatter correction accuracy while dramatically reducing computational requirements and processing time.
3Reliability
If simple thresholding is used to remove scattered photons, then the processing is easier, but the reliability deteriorates due to measurement uncertainties
Solution Approach 1:
The patent implements feedback by using the object's attenuation map (obtained from CT or transmission measurements) to inform the scatter estimation process. This feedback loop allows the scatter correction to adapt to the specific anatomical structure being imaged, improving reliability by accounting for actual photon paths through different tissue densities rather than using fixed thresholds.
Solution Approach 2:
The patent replaces the mechanical thresholding approach with a physics-based computational model for scatter estimation. Instead of simply discarding photons below an energy threshold, the system uses attenuation information to model and subtract scatter contributions, providing more reliable correction that accounts for the actual physical processes occurring during imaging.
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 significantly improves the accuracy of image generation by effectively removing scatter radiation modifications, resulting in a more reliable and precise graphical representation of the radioactive imaging agent distribution within the body, reducing errors associated with photon scattering.
Implementation Method 1
the radioactive imaging agent is designed for being distributed within the body of the subject and for leading to an emission of two photons of a well-defined energy of 511 keV per radioactive decay
Implementation Method 2
Considering that each emitted photon travels along a straight line
Implementation Method 3
due to an interaction between the emitted photons and matter, including but not limited to the body of the subject and the detector equipment, photon scattering may occur at one or more scattering objects, which results in a change of the direction of at least one of the travelling photons
Implementation Method 4
the PET detector is, therefore, designed for detecting coincident pairs of the emitted photons, also denoted as 'coincidences.' This kind of information as detected by the PET detector is, subsequently, used as an input for image generation
Data Source
AI summary
A method for training of a process for generating an image of an object from measurement data modified by scatter radiation. Individual sets of measured data are each made up of matrix elements. Each matrix element corresponds to an individual detector element that detects ionizing radiation. Signals measured by the individual detector elements in an energy bin are assigned as values for each matrix element. The individual sets of measured data and a template image are used as input for a procedure for determining a correction image for correcting a modification of the measured data by the scatter radiation. The preliminary image obtained using the individual sets of measured data is adjusted to the template. These steps are repeated until the deviation between the preliminary image and the template image is below a threshold. The procedure is used to generate the image of the object from the measurement data.