Neural ACF Map Generation From PET Emission Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing attenuation correction methods in multimodality medical imaging, such as PET/CT and PET/MR, face challenges with tissue misclassification, truncation, and incomplete bone atlas addition, leading to inaccurate reconstructed images, especially in application-specific and population-specific protocols.
Innovation Solution
A framework that utilizes deep artificial neural networks to directly generate attenuation correction factor (ACF) maps from raw emission data, bypassing the need for image reconstruction, using convolutional neural networks (CNNs) to synthesize ACF maps in PET data projection space.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional attenuation correction methods (CT-based or MR-based segmentation) are used, then attenuation maps can be generated, but tissue misclassification, truncation, and incomplete bone atlas addition occur leading to inaccurate reconstructed images
Solution Approach 1:
The patent replaces traditional mechanical/image-based attenuation correction methods (CT-based Hounsfield unit conversion or MR-based segmentation) with a neural network-based system. The neural network directly processes PET emission data to generate attenuation correction factors, eliminating the need for separate CT or MR segmentation processes and avoiding the associated tissue misclassification and truncation errors.
Solution Approach 2:
The patent introduces a neural network as an intermediary between the PET emission data and the attenuation correction process. Instead of directly converting CT or MR images to attenuation maps, the neural network acts as a mediator that learns the complex relationship between emission data patterns and attenuation characteristics, producing accurate correction factors without traditional imaging artifacts.
2Measurement precision
If CT scans are acquired for PET/CT imaging to generate attenuation maps, then photon-attenuation effects can be corrected, but additional imaging time, radiation exposure, and processing complexity are required
Solution Approach 1:
The patent extracts the attenuation correction function from the traditional PET/CT or PET/MR imaging protocol. Instead of requiring separate CT or MR scans to be acquired and processed, the system extracts attenuation correction factors directly from the PET emission data itself using neural networks, eliminating the need for additional anatomical imaging modalities.
Solution Approach 2:
The patent makes the PET emission data serve multiple functions: both functional imaging and attenuation correction. The same PET scan data that provides functional information is also used to generate the attenuation correction factors, eliminating the need for separate CT or MR scans and simplifying the overall imaging protocol.
3Measurement precision
If attenuation correction is performed using traditional methods, then quantitatively accurate reconstructed images can be obtained, but significant processing time and computational resources are required
Solution Approach 1:
The patent performs preliminary learning during the training phase, where the neural network is pre-trained on large datasets of emission data and corresponding attenuation maps. This preliminary action allows the network to capture complex attenuation patterns and relationships, enabling fast inference during actual image reconstruction without requiring time-consuming computational processes.
Solution Approach 2:
The patent replaces computationally intensive traditional attenuation correction algorithms (based on CT Hounsfield unit conversions or MR segmentation) with a neural network model that has learned the complex relationships during training. The neural network's parameterized functions enable rapid inference, significantly reducing processing time while maintaining quantitative accuracy.
Data Source
AI summary
A framework for medical image data processing. An attenuation correction factor (ACF) map is generated by applying the raw emission data directly to one or more artificial neural networks. A medical image may then be reconstructed from the ACF map.