Composite PET-CT Image Generation via Deep Learning Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current PET/CT imaging methods require additional CT scans, increasing radiation exposure and costs, and are prone to inaccuracies due to manual data registration and patient movement, which complicates attenuation correction and image quality.
Innovation Solution
A deep learning-based method using a two-stage generative adversarial network to generate composite PET/CT images from non-attenuation-corrected PET images, eliminating the need for additional CT scans by learning mapping relationships between PET and CT images, thereby reducing radiation doses and improving image accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If additional CT scan is performed for attenuation correction, then anatomical information for locating lesion positions is provided, but patient radiation exposure increases
Solution Approach 1:
The patent creates a pseudo-CT image as a copy substitute for the real CT image. The pseudo-CT is generated from MRI data through deep learning, replicating the attenuation correction function of CT without requiring actual CT radiation. This copying approach maintains the necessary anatomical information while eliminating harmful radiation exposure.
Solution Approach 2:
The patent introduces MRI as an intermediary modality between PET and CT. Instead of directly using CT for attenuation correction, the system uses MRI to generate pseudo-CT images that serve as the intermediary for PET attenuation correction. This intermediary approach provides anatomical information while avoiding CT radiation.
2Measurement precision
If MRI sequences are acquired for generating pseudo-CT, then attenuation correction is achieved, but examination time increases
Solution Approach 1:
The patent performs preliminary action by acquiring MRI data during the PET/MRI examination process before the deep learning processing is needed. The MRI sequences are acquired in advance, and the deep learning model processes this pre-acquired data to generate pseudo-CT images, thereby avoiding additional time-consuming scanning procedures.
Solution Approach 2:
The patent replaces the mechanical scanning process with deep learning computation. Instead of using time-consuming iterative reconstruction methods or manual processing, the system uses a pre-trained deep learning model to rapidly generate pseudo-CT images from MRI data, substituting computational intelligence for mechanical scanning time.
3Measurement precision
If manual data registration is performed for PET/MRI compositing, then anatomical alignment is achieved, but registration accuracy is compromised by patient movement
Solution Approach 1:
The patent applies self-service by enabling the deep learning model to automatically perform registration and compositing without manual intervention. The model learns the transformation relationships between PET, MRI, and pseudo-CT images during training, and automatically applies these transformations during inference, eliminating manual registration errors and reducing sensitivity to patient movement.
Solution Approach 2:
The patent incorporates feedback mechanisms through the deep learning training process. The model learns from paired training data (PET-MRI-CT triplets) and continuously improves its registration and compositing accuracy. The loss function provides feedback on alignment quality, enabling the system to automatically adjust and optimize registration parameters without manual intervention.
Data Source
AI summary
The present disclosure discloses a method and a system for generating a composite PET-CT image based on a non-attenuation-corrected PET image. The method includes: constructing a first generative adversarial network and a second generative adversarial network; obtaining a mapping relationship between a non-attenuation-corrected PET image and an attenuation-corrected PET image by training the first generative adversarial network; obtaining a mapping relationship between the attenuation-corrected PET image and a CT image by training the second generative adversarial network; and generating the composite PET-CT image by utilizing the obtained mapping relationships. According to the present disclosure, a high-quality PET-CT image can be directly composited from a non-attenuation-corrected PET image, and medical costs can be reduced for patients, and radiation doses applied to the patients in examination processes can be minimized.