ECG AI Model Updating With Reliability-Based Data Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing artificial intelligence models for electrocardiogram reading may suffer from reduced performance when re-trained with unreliable electrocardiogram data, leading to potential overloading and instability in electrocardiogram reading services.
Innovation Solution
A method to update the AI model by classifying electrocardiogram data based on relative value using multiple determinations, re-training with high-value data, and selecting the model with the highest accuracy to ensure reliable and efficient model performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If electrocardiogram data is used to re-train an artificial intelligence model, then the model performance may be improved, but the system stability may be compromised when unreliable data is used
Solution Approach 1:
The patent changes the parameter of data selection from random or comprehensive data usage to selective data usage based on reliability assessment. By introducing reliability determination as a parameter change criterion, the system selects only high-reliability electrocardiogram data for model re-training, thus improving model performance while maintaining system stability.
Solution Approach 2:
The system performs self-assessment of data reliability through automated determination processes. The electrocardiogram reading service system itself evaluates the reliability of its own data before using it for model re-training, eliminating the need for external manual verification and enabling autonomous quality control.
2Measurement precision
If all cumulative electrocardiogram data is used for model re-training, then comprehensive training is achieved, but the processing time and computational load increase significantly
Solution Approach 1:
The patent extracts only the essential and high-value subset of electrocardiogram data from the cumulative dataset for model re-training. By determining data reliability and selecting only high-reliability data, the system extracts the most valuable training samples, reducing processing time and computational load while maintaining or improving model accuracy.
Solution Approach 2:
Instead of using all available cumulative data (excessive action), the patent applies partial action by selectively using only high-reliability electrocardiogram data for model re-training. This partial approach reduces the training dataset size significantly, decreasing processing time and computational resources required while still achieving effective model improvement.
3Productivity
If newly obtained electrocardiogram data is used for model re-training without reliability verification, then rapid model updates are possible, but the model performance cannot be guaranteed
Solution Approach 1:
The patent performs preliminary reliability determination on newly obtained electrocardiogram data before using it for model re-training. This preliminary assessment action ensures that only high-reliability data is selected for training, guaranteeing model performance improvement while maintaining rapid update capability. The reliability check is performed in advance, enabling quick identification of suitable training data.
Data Source
AI summary
The present disclosure is directed to a method, program, and device of updating an artificial intelligence model for electrocardiogram reading. The method is performed by a computing device including at least one processor. The method includes: obtaining electrocardiogram data; classifying the relative value of the electrocardiogram data using at least one of a first value determination, a second value determination, and a third value determination based on preset reference value ranges by analyzing the electrocardiogram data; and combining pieces of electrocardiogram data whose relative values have been classified to re-train a previously constructed first machine learning model and to train one or more new second machine learning models.