PET Image Reconstruction Using Differentiable Inverse Rendering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current PET image reconstruction methods face challenges with noise, artifacts, partial volume effects, and overfitting, and deep learning-based methods require large datasets and struggle with generalization across different tracers.
Innovation Solution
A physically-based differentiable rendering (PBDR) approach for PET image reconstruction that uses forward and inverse rendering processes, incorporating Monte Carlo simulations, auto-differentiation, and gradient-based optimization to iteratively refine pixel values and account for physical processes like Compton scatter and attenuation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If traditional PET image reconstruction methods are used, then the reconstruction process is computationally simpler and faster, but the image resolution is lower and partial volume effects are more pronounced
Solution Approach 1:
The patent replaces traditional mechanical iterative reconstruction algorithms with a differentiable rendering system that uses gradient-based optimization. The forward rendering process simulates photon transport through the imaged object using differentiable physics models, and gradients are automatically computed via backpropagation to update the 3D scene representation, substituting conventional iterative mathematical reconstruction methods.
Solution Approach 2:
The patent introduces a differentiable forward rendering process as an intermediary between the 3D scene representation and the measured projection data. This rendering process acts as a differentiable operator that simulates the physics of photon transport, enabling gradient computation while maintaining physical accuracy, thus bridging the gap between simple scene representation and complex measurement data.
2Manufacturing precision
If deep learning-based reconstruction methods are used, then image quality can be improved, but large datasets are required and generalization across different tracers is difficult
Solution Approach 1:
The patent changes the fundamental parameters of the reconstruction approach by using a physics-based differentiable rendering model instead of data-driven deep learning models. The system optimizes physical parameters such as attenuation coefficients, scatter coefficients, and scene geometry directly from projection data using gradient descent, eliminating the need for large training datasets and enabling generalization across different tracers and imaging scenarios.
Solution Approach 2:
The patent enables the reconstruction system to be self-sufficient by using the measured projection data itself to train and optimize the 3D scene representation through gradient-based optimization. The differentiable rendering process allows the system to learn from the data without requiring external training datasets, making the method adaptable to different tracers and imaging conditions.
3Measurement precision
If more photons are detected to reduce noise, then the signal-to-noise ratio improves, but the scan time increases
Solution Approach 1:
The patent performs preliminary action by using the differentiable forward rendering process to simulate and account for noise and physical effects before actual image reconstruction. The system models photon transport, attenuation, and scatter in advance, allowing gradient-based optimization to effectively utilize the measured data and achieve high signal-to-noise ratio images from limited photon counts without requiring extended scan times.
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
Enhances image resolution and reduces partial volume effects, providing higher quality training data for deep learning methods and improving spatial resolution and signal-to-noise ratio in PET imaging.
Implementation Method 1
The physical processes may comprise a combination of Compton scatter, attenuation, and/or transmission
Implementation Method 2
The physical processes may comprise a combination of Compton scatter, attenuation, and/or transmission
Implementation Method 3
The simulation may account for statistical interactions of photons with a medium along each LOR
Data Source
AI summary
The present disclosure relates to reconstructing positron emission tomography (PET) images. The approach may involve receiving measured sinogram data from one or more scans by a PET scanner. The measured sinogram data may represent measured projections from the PET scanner. The approach may involve performing forward rendering to generate a rendered sinogram. Forward rendering may comprise sampling a number of positions corresponding to each crystal detector in a plurality of crystal detectors. The positions may define lines of response (LORs) between crystal pairs. The approach may involve performing inverse rendering based on the measured sinogram data and the rendered sinogram. Inverse rendering may comprise applying auto-differentiation for gradient-based optimization. The rendering may be performed iteratively to update pixel values of an emission image until a stopping criterion. A reconstructed PET image based on the updated emission image may be output following the stopping criterion being satisfied.