Ultrasound Training Data Standardization for AI Model Adaptability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning models for ultrasound diagnostics face challenges in handling time-varying image data due to variations in sweep speed and heart rate, requiring extensive training data for each condition, which increases costs and complexity.
Innovation Solution
Standardizing time-varying image data to a predetermined sweep speed and heart rate allows for the generation of training data that can be used across multiple conditions, reducing the need for extensive data collection and enabling efficient training of machine learning models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If training data is prepared for each sweep speed condition, then the AI model can handle different sweep speeds, but the amount of training data required increases significantly
Solution Approach 1:
The patent applies parameter changes by transforming time-varying image data through time-axis scaling operations. The data processing unit scales the time axis of training data to match different sweep speed conditions, allowing a single base training dataset to be adapted for multiple sweep speed scenarios. This resolves the contradiction by changing the temporal parameter of the data rather than creating separate datasets for each condition.
Solution Approach 2:
The patent implements universality by creating a standardized training data format that can serve multiple sweep speed conditions. The time-varying image data is processed into a universal representation that can be scaled to match any sweep speed requirement, making the training data multi-functional across different operating conditions without requiring separate specialized datasets for each sweep speed.
2Adaptability or versatility
If training data is prepared for each heart rate condition, then the AI model can handle different heart rates, but the amount of training data required increases significantly
Solution Approach 1:
The patent applies parameter changes by scaling the time axis of training data to accommodate different heart rate conditions. The data processing unit transforms the temporal characteristics of the training data to match various heart rates, allowing a single base dataset to be adapted for multiple heart rate scenarios through parameter transformation rather than data multiplication.
Solution Approach 2:
The patent implements universality by creating a standardized training data format that can serve multiple heart rate conditions. The time-varying image data is processed into a universal representation that can be scaled to match any heart rate requirement, making the training data multi-functional across different physiological conditions without requiring separate specialized datasets for each heart rate.
3Measurement precision
If separate training data is collected for each combination of sweep speed and heart rate, then the AI model achieves high accuracy across all conditions, but the complexity and cost of data preparation increases
Solution Approach 1:
The patent applies parameter changes by implementing automated time-axis scaling transformations on training data. The data processing unit systematically adjusts the temporal parameters of training data to match different sweep speed and heart rate conditions, maintaining detection accuracy across all scenarios while eliminating the need for manual data collection and preparation for each condition combination.
Solution Approach 2:
The patent implements preliminary action by pre-processing training data into a standardized format with normalized time axes. This preliminary transformation allows the training data to be readily scaled and adapted to any sweep speed or heart rate condition without requiring complex post-processing or re-collection, thereby reducing overall data preparation complexity while maintaining accuracy.
Data Source
AI summary
Techniques for efficiently generating time-varying image data for use in training a machine learning model are disclosed. An aspect of the present disclosure relates to a machine learning model trained using training data that includes at least one piece of training time-varying image data of second time-varying image data and third time-varying image data, the second time-varying image data being obtained by standardizing first time-varying image data in a time direction, the first time-varying image data being based on a reception signal for image generation received by an ultrasound probe, third time-varying image data being based on the second time-varying image data, and, and ground truth data including a detection target corresponding to the at least one piece of training time-varying image data.