Residual Inception Encoder-Decoder Network for Amyloid PET Harmonization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current amyloid and tau PET imaging methods using different tracers face challenges due to inconsistent amyloid positivity thresholds and inter-tracer variability, which complicates multi-center studies and longitudinal tracking of amyloid accumulation.
Innovation Solution
A deep learning-based method utilizing a residual inception encoder-decoder network (RIED-Net) is employed to harmonize PET images across different tracers, converting original PET images into target PET images that simulate the use of a different tracer, thereby standardizing amyloid measurements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If multiple different PET tracers are used for amyloid imaging, then more research centers can participate and imaging capacity increases, but inter-tracer variability leads to inconsistent amyloid positivity thresholds and measurement discrepancies
Solution Approach 1:
The patent applies parameter changes by transforming PET images from one tracer's parameter space to another tracer's parameter space using deep learning. The model learns tracer-specific parameters (binding affinity, kinetic behavior, non-specific binding characteristics) and adjusts the image data accordingly, converting images to match the target tracer's measurement characteristics and resolve threshold inconsistencies
Solution Approach 2:
The patent introduces a deep learning model as an intermediary between different PET tracers. This intermediary learns the mapping relationships between tracers and can convert images from any tracer to any other tracer, acting as a mediator that enables consistent comparison across different imaging modalities without requiring direct paired training data
2Reliability
If tracer-specific imaging protocols and scanners are used at different centers, then imaging quality is optimized for each center's equipment, but scanner differences and protocol variations increase inter-site variability
Solution Approach 1:
The patent creates a universal deep learning model that can process and harmonize images from multiple different scanners and protocols. The model is trained on diverse data from various centers and equipment, enabling it to function universally across different imaging systems and eliminate scanner-specific artifacts and protocol variations
3Device complexity
If traditional quantification methods are used without harmonization, then the analysis pipeline is simple, but amyloid measurements show significant variability across different tracers and pipelines
Solution Approach 1:
The patent replaces traditional mechanical/image processing harmonization methods with a deep learning-based computational approach. Instead of using conventional image registration, normalization, and manual correction techniques, the system uses neural networks to automatically learn and correct tracer-specific variations, achieving better harmonization with automated processing
Data Source
AI summary
Using deep learning, imaging harmonization among images produced with differing PET tracers is achieved. A method may include providing an original PET image of the brain using an original PET tracer, providing a target PET tracer, and converting, using a deep learning neural network, the original PET image into a target PET image simulating the image that would be obtained had the target PET tracer been used.