PET Histo-Image Super Time-of-Flight Resolution Using Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current PET scanners have limited time resolution, with low-cost systems providing around 600 ps response time and high-end systems up to 200 ps, while scanners achieving 10 ps are theoretical, hindering improved image resolution and clinical outcomes.
Innovation Solution
A system utilizing a trained neural network to enhance the resolution of PET images by back-projecting low-resolution PET data, generating improved-resolution histo-images, and reconstructing PET images with higher detail through a neural network-based reconstruction process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional PET scanners are used, then the system is cost-effective and readily available, but the time resolution is limited to 600-200 ps
Solution Approach 1:
The patent creates a virtual copy of the PET imaging system at a higher resolution level. A neural network is trained to take low-resolution histo-images (from conventional 600-200 ps scanners) and generate high-resolution simulated histo-images (10 ps equivalent). This allows the expensive high-resolution system to be replicated computationally without requiring the expensive hardware, thus improving measurement precision while avoiding the full cost and complexity of a high-resolution physical scanner.
Solution Approach 2:
The patent replaces the mechanical/hardware solution for improving time resolution with a computational/software solution. Instead of building a physically faster scanner (which would cost more and be more complex), the system uses neural network algorithms to simulate and enhance image resolution. The neural network learns the relationship between low-resolution input images and high-resolution output images, substituting the need for expensive hardware improvements with computational processing.
2Measurement precision
If neural network-based resolution enhancement is applied, then image resolution improves from 600 ps to 200 ps or 10 ps, but computational processing complexity increases
Solution Approach 1:
The neural network is pre-trained using a training dataset before actual imaging. During training, the network learns the mapping from low-resolution to high-resolution images by analyzing many examples. This preliminary training action encapsulates the complex computational logic into the network's weights and architecture. During actual use, only the forward pass through the trained network is needed, which is much simpler than the training process itself.
Solution Approach 2:
The neural network acts as an intermediary between the low-resolution input data and the high-resolution output images. Instead of directly processing raw detector signals to achieve high resolution (which would be computationally intensive), the system first creates low-resolution histo-images through conventional reconstruction, then uses the neural network as an intermediate step to transform these into high-resolution images. This intermediary approach breaks down the complex computational task into manageable stages.
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
The system achieves PET image resolution enhancement from 600 ps to 200 ps or even 10 ps, significantly improving image detail and clinical analysis capabilities using existing PET systems.
Implementation Method 1
Time-of-flight (TOF) PET measures the difference between the detection times of the two gamma photons arising from the annihilation event. This difference can be used to estimate a particular position along the LOR at which the annihilation event occurred.
Implementation Method 2
Radioactive decay of the radionuclide generates positrons, which eventually encounter electrons and are annhilated thereby. The annihilation event produces two gamma photons that travel in approximately opposite directions.
Implementation Method 3
input the first histo-image to a trained neural network, receive a second histo-image from the trained neural network
Implementation Method 4
back-project the first PET dataset to generate a first histo-image
Data Source
AI summary
Systems and methods of generating improved resolution histo-images are disclosed. A system includes a positron emission tomography (PET) imaging modality configured to execute a first scan to acquire a first PET dataset and a processor configured to back-project the first PET dataset to generate a first histo-image having a first resolution, input the first histo-image to a trained neural network, receive a second histo-image from the trained neural network, and input the second histo-image to a reconstruction process configured to generate a reconstructed PET image. The second histo-image has a second resolution higher than the first resolution. The second histo-image represents the first PET dataset.