Image Processing Model Training for Rare Tracers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies face challenges in training image processing models for ECT images generated using tracers that are not commonly used, due to the scarcity of sufficient training data.
Innovation Solution
A system and method for model training that obtains training samples including images from both commonly and less commonly used tracers, and generates a trained image processing model by training a preliminary model using these samples, thereby overcoming the scarcity of data for less commonly used tracers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a trained model is trained using sample images generated based on a specific tracer, then the model can process images for that tracer, but it cannot be used to process images generated based on other tracers
Solution Approach 1:
The patent applies universality by training a single image processing model using a composite dataset that includes sample images from multiple different tracers (e.g., 18F-FDG, 68Ga-PSMA, 11C-MET). This allows the model to serve multiple functions and process images generated from various tracers, rather than requiring separate models for each tracer type.
Solution Approach 2:
The patent merges training data from different tracer types into a unified training dataset. By combining sample images and reference images from multiple tracers, the system creates a comprehensive training set that enables the model to learn common image processing patterns across different tracer modalities, resolving the limitation of tracer-specific models.
2Manufacturing precision
If sufficient training data is obtained for a rarely used tracer, then a high-quality model can be trained, but it is difficult to obtain sufficient training data for tracers that are not commonly used
Solution Approach 1:
The patent combines limited training data from rare tracers with abundant training data from common tracers into a unified dataset. This merging approach allows the model to leverage the large volume of data from commonly used tracers (like 18F-FDG) to compensate for the scarcity of data from rarely used tracers (like 68Ga-PSMA), thereby achieving sufficient training data quantity for high-quality model training.
Solution Approach 2:
The trained model achieves universal applicability across both common and rare tracers. By training on a diverse dataset that includes both frequently used and rarely used tracer images, the model learns generalized image processing capabilities that work effectively across all tracer types, ensuring high model quality even for tracers with limited training data.
3Reliability
If separate models are trained for each tracer type, then each model can be optimized for its specific tracer, but the complexity of maintaining multiple models increases
Solution Approach 1:
The patent implements a single universal image processing model that can handle multiple tracer types, eliminating the need to maintain separate models for each tracer. This reduces system complexity while maintaining processing accuracy through comprehensive training on diverse tracer data, including both common and rare tracers.
Solution Approach 2:
The patent merges the functionality of multiple tracer-specific models into a single unified model. By combining training data from various tracers and training one model on this composite dataset, the system achieves the processing capabilities of multiple specialized models while reducing the complexity of model maintenance and deployment.
Data Source
AI summary
The present disclosure provides a system and method for image reconstruction. The method may include obtaining training samples, the training samples including at least one sample first image generated based on a first tracer, at least one reference first image each of which corresponds to one of the at least one sample first image and has a higher image quality than the corresponding sample first image, at least one sample second image generated based on a second tracer different from the first tracer, and at least one reference second image each of which corresponds to one of the at least one sample second images and has a higher image quality than the corresponding sample second image; and generating a trained image processing model by training a preliminary model using the training samples.