Medical Imaging Model Training with Synthetic High-Quality Scan Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for obtaining high-quality training data in medical imaging using machine learning models involve prolonged scanning times, increased radiation doses, and susceptibility to motion artifacts, which are undesirable for subjects.
Innovation Solution
A system and method for generating high-quality training data by processing preliminary training data through operations such as data splitting, rebinning, and down-sampling, followed by training a machine learning model to optimize scanning data quality parameters like SNR, spatial resolution, and image contrast.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If scanning time is prolonged to obtain high-quality training data, then data quality improves, but subject discomfort increases and motion artifacts increase
Solution Approach 1:
The patent creates synthetic training target data by copying and transforming existing preliminary training data through data splitting, rebinning, and down-sampling operations. This allows the machine learning model to learn from synthesized high-quality data without requiring actual prolonged scanning, thereby avoiding motion artifacts while maintaining data quality improvement.
Solution Approach 2:
The patent performs data processing operations (splitting, rebinning, down-sampling) on preliminary training data in advance to generate training input data before model training. This preliminary preparation of training data enables the model to learn quality enhancement without requiring actual prolonged scanning of subjects, thus preventing motion artifacts.
2Measurement precision
If scanning time is prolonged to obtain high-quality training data, then data quality improves, but subject discomfort increases
Solution Approach 1:
The patent synthesizes training target data by copying and transforming existing preliminary training data through processing operations. This allows the system to create high-quality training data without requiring actual prolonged scanning of subjects, thereby improving data quality while avoiding the time loss associated with extended scanning procedures.
Solution Approach 2:
The patent replaces the mechanical process of prolonged physical scanning with a computational data processing system. Instead of physically extending scan time to improve data quality, the system uses data splitting, rebinning, and down-sampling operations to synthesize high-quality training data, substituting mechanical scanning extension with information processing.
3Measurement precision
If radiation dose is increased to obtain high-quality training data, then data quality improves, but harmful radiation exposure increases
Solution Approach 1:
The patent synthesizes training target data by copying and transforming existing preliminary training data through data processing operations. This allows the system to create high-quality training data without requiring actual increased radiation dose scanning, thereby improving data quality while avoiding additional harmful radiation exposure to subjects.
4Measurement precision
If advanced scanners with higher TOF sensitivity are used to obtain high-quality training data, then data quality improves, but device complexity and cost increase
Solution Approach 1:
The patent synthesizes training target data by copying and transforming existing preliminary training data from scanners with lower TOF sensitivity through data processing operations. This allows the system to create training data that simulates the quality of advanced scanner output without requiring actual advanced scanners, thereby improving effective data quality while avoiding increased device complexity and cost.
Solution Approach 2:
The patent changes data parameters through processing operations (splitting, rebinning, down-sampling) to transform preliminary training data into synthesized training target data with enhanced quality characteristics. This parameter transformation approach allows the system to achieve data quality equivalent to advanced scanners without requiring actual hardware upgrades, avoiding increased device complexity.
Data Source
AI summary
The present disclosure provides a medical imaging system and method. The method may include obtaining a machine learning model and preliminary training data of at least one sample subject.The method may also include generating training input data by processing the preliminary training data, the preliminary training data being superior to the training input data with respect to a data quality parameter. The method may further include determining a trained machine learning model by training the machine learning model based on the training input data and the preliminary training data, the preliminary training data being configured as training target data of the machine learning model.