PET Image Reconstruction Using Long-Axis Scanner Training Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for obtaining high-quality training data in medical imaging, such as PET scans, often require prolonged scanning times, increased radiation doses, and are susceptible to motion artifacts, leading to discomfort for the subject and limited improvement in image quality.
Innovation Solution
A machine learning model is trained using high-quality preliminary training data, processed to generate optimized training input data, allowing for improved image quality without extending scan time or increasing radiation dose by utilizing a long-axis PET scanner's data to enhance regular PET scanners.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If scanning time is prolonged to obtain high-quality training data, then image quality is improved, but subject discomfort increases and productivity decreases
Solution Approach 1:
The patent applies preliminary action by pre-training a machine learning model using high-quality training data obtained from prolonged scanning. This pre-trained model is then deployed to enhance quality of routine fast scans, allowing high-quality images to be produced without actually performing prolonged scans during clinical operation.
Solution Approach 2:
The patent introduces a machine learning model as an intermediary between the scanning system and the final image output. This model acts as a mediator that can transform standard-quality scan data into high-quality images through learned transformations, bypassing the need for prolonged scanning.
2Measurement precision
If radiation dose is increased to obtain high-quality training data, then image quality is improved, but harmful factors to the subject increase
Solution Approach 1:
The patent performs the radiation exposure and data collection in advance during the offline training phase. The trained model can then generate high-quality images from standard-dose scans without requiring additional radiation exposure during clinical use, effectively separating the quality enhancement process from radiation exposure.
Solution Approach 2:
The patent creates a computational copy of the high-quality imaging process through the trained machine learning model. Instead of repeatedly exposing subjects to high radiation doses, the model learns from such data and reproduces the quality enhancement effect on standard-dose scans.
3Measurement precision
If conventional methods are used to improve training data quality, then image quality is improved, but device complexity and operational difficulty increase
Solution Approach 1:
The patent replaces complex physical modifications to the scanning system with a software-based machine learning model. Instead of modifying hardware to achieve higher quality, the system uses computational methods to enhance images from standard scanners, significantly reducing system complexity.
Solution Approach 2:
The patent changes the approach from modifying physical scanning parameters (time, dose) to modifying data processing parameters through machine learning. The model learns optimal transformations and applies them to enhance image quality without changing the fundamental scanning parameters.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A medical imaging system (100) implements a method. The method includes: obtaining a machine learning model and preliminary training data of at least one sample subject (510), generating training input data by processing the preliminary training data (520), the preliminary training data being superior to the training input data with respect to a data quality parameter, determining a trained machine learning model by training the machine learning model based on the training input data and the preliminary training data (530), the preliminary training data being configured as training target data of the machine learning model.