AI-Driven PET Motion Correction Without Motion Phase Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing molecular imaging (MI) techniques struggle to generate diagnostically-suitable images efficiently due to patient motion, which causes organ displacement and blurring, and current motion correction methods are computationally intensive and require accurate characterization of motion phases.
Innovation Solution
A neural network is trained using anatomical and motion-corrected PET images to directly generate a motion-corrected PET image from a CT image, eliminating the need for motion identification and vector computation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If traditional motion correction methods are used (segmenting emission data into motion-frozen frames or incorporating motion patterns into reconstruction), then motion artifacts are reduced, but computational complexity and processing time increase significantly
Solution Approach 1:
The patent replaces complex computational motion correction algorithms with a trained neural network model. The neural network learns motion patterns during training and automatically corrects motion artifacts during inference, substituting iterative computational methods with a direct neural network-based approach that requires less computational resources during actual image correction.
Solution Approach 2:
The patent performs preliminary training of the neural network model using motion-corrected reference images before actual PET image correction. During this training phase, the system learns the relationship between motion-distorted images and their corrected versions, so that during actual use, the pre-trained network can quickly correct new images without requiring complex real-time motion analysis and reconstruction.
2Measurement precision
If accurate motion phase characterization is performed to generate correction factors, then motion correction accuracy improves, but processing time and computational resources increase
Solution Approach 1:
The patent substitutes the traditional workflow of motion phase detection, characterization, and correction factor calculation with a neural network that directly processes the PET image data. The neural network internally learns and applies appropriate corrections based on patterns it discovered during training, eliminating the need for explicit motion phase characterization and correction factor computation.
Solution Approach 2:
The patent uses a trained neural network model that has learned from training data containing motion-corrected reference images. The network copies the corrective transformations learned during training and applies them to new PET images, replacing the need for real-time motion analysis and correction factor generation.
3Measurement precision
If external devices or raw data analysis are used to monitor motion patterns, then motion detection accuracy improves, but system complexity and resource requirements increase
Solution Approach 1:
The patent extracts and removes the need for external motion monitoring devices and complex motion analysis systems. Instead of using separate systems to detect and characterize motion, the neural network is trained to recognize motion patterns directly from the PET image data itself, eliminating the need for external sensors and complex motion monitoring infrastructure.
Solution Approach 2:
The patent enables the PET imaging system to correct its own motion artifacts using the neural network. The network is trained on PET image data and learns to identify and correct motion distortions directly from the imaging data itself, without requiring external motion monitoring systems or additional hardware to detect and characterize patient motion.
Data Source
AI summary
Systems and methods include acquisition of an anatomical image of an object, acquisition of molecular imaging data of the object at the plurality of photon detectors, reconstruction of a functional image based on the molecular imaging data, input of the anatomical image and the functional image to a trained neural network to generate a second functional image, and presentation of the second functional image.