PET Scatter Sinogram Estimation Using DCNN Reconstruction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current scatter correction methods in PET imaging, such as Monte Carlo simulations and model-based simulations, are either too slow for real-time applications or computationally expensive, and existing deep convolutional neural network (DCNN) approaches struggle to directly estimate scatter from emission and attenuation sinograms due to the lack of a direct relationship between these sinograms and scatter sinograms.
Innovation Solution
A machine learning-based system using a DCNN is trained to estimate scatter directly from nuclear medicine images, utilizing Monte Carlo or model-based scatter correction methods for training, and produces scatter sinograms or scatter-corrected images, employing a U-Net-like structure and minimizing a loss function for accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If Monte Carlo simulations are used for scatter correction, then accuracy is improved, but computational speed deteriorates
Solution Approach 1:
The patent creates a trained DCNN model that copies the scatter correction capabilities of Monte Carlo simulations. The neural network is trained using Monte Carlo simulation data but executes much faster during inference, effectively copying the accurate scatter correction behavior without the computational burden of real-time Monte Carlo simulations.
Solution Approach 2:
The patent replaces the mechanical/computational physics-based Monte Carlo simulation system with a machine learning-based DCNN system. This substitution maintains the accuracy of scatter correction while dramatically improving computational speed through the neural network's efficient pattern recognition and prediction capabilities.
2Loss of time
If model-based scatter simulations are used, then computational time is reduced, but accuracy deteriorates due to inability to handle multiple scattered events
Solution Approach 1:
The DCNN model copies the comprehensive scatter correction performance of full Monte Carlo simulations (which handle multiple scattered events) while operating at the speed of model-based methods. The network learns to predict scatter sinograms that accurately represent multiple scattered events from training data generated by Monte Carlo simulations.
Solution Approach 2:
The patent changes the approach from physics-based modeling parameters to data-driven learning parameters. By training the DCNN on Monte Carlo simulation outputs, the network captures complex scatter behavior through learned parameters rather than explicit physical models, enabling accurate prediction of multiple scattered events without the computational cost of traditional modeling.
3Speed
If DCNN estimates scatter directly from emission and attenuation sinograms, then speed is improved, but reliability deteriorates due to lack of direct relationship
Solution Approach 1:
The patent introduces an intermediary approach by training the DCNN on data that does directly represent the physical relationship between emission/attenuation sinograms and scatter sinograms. The network learns the transformation rules from the training data, establishing a reliable mapping that respects the physical constraints while maintaining computational speed.
Solution Approach 2:
The DCNN copies the reliable physical relationships from the training data (generated by Monte Carlo simulations) to maintain accuracy during inference. The network preserves the causal relationships between emission/attenuation sinograms and scatter sinograms by learning from authentic physical data rather than making unsubstantiated assumptions.
4Measurement precision
If second order scattered events are modeled in model-based simulations, then accuracy is improved, but device complexity increases
Solution Approach 1:
The DCNN copies the high accuracy of second-order scatter modeling without replicating the complex iterative calculations. The network encapsulates the complexity of second-order scatter physics in its trained weights, providing accurate predictions through simple forward propagation rather than complex computational algorithms.
Solution Approach 2:
The patent replaces the complex mechanical/mathematical system of second-order scatter modeling with a neural network system. The substitution maintains the accuracy of modeling multiple scattered events while eliminating the computational complexity through data-driven learning and efficient neural network inference.
Data Source
AI summary
A method, system, and computer readable medium to perform nuclear medicine scatter correction estimation, sinogram estimation and image reconstruction from emission and attenuation correction data using deep convolutional neural networks. In one embodiment, a Deep Convolutional Neural network (DCNN) is used, although multiple neural networks can be used (e.g., for angle-specific processing). In one embodiment, a scatter sinogram is directly estimated using a DCNN from emission and attenuation correction data. In another embodiment a DCNN is used to estimate a scatter-corrected image and then the scatter sinogram is computed by a forward projection.