Learning-Model Blood Glucose Estimation from Biosignals
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing blood glucose monitoring systems face limitations in accurately detecting hypoglycemic conditions and require extensive manual configuration, leading to increased workload and cost, while continuous glucose monitoring systems (CGMS) are cumbersome and inefficient in processing biosignals.
Innovation Solution
A blood glucose measurement method using a learning model that automates the process of deriving a blood glucose value from biosignals by training the model with biometric, blood glucose-related, environmental, and health data, utilizing a neural network structure with multiple hidden layers and varied activation functions to improve accuracy and simplify system configuration.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional functional configurations (sampling, noise removal, filtering, calibration) are used to process biosignals, then blood glucose measurement can be achieved, but the amount of computation becomes very large and the system requires extensive manual configuration
Solution Approach 1:
The patent replaces the conventional mechanical signal processing system (sampling, noise removal, filtering, calibration) with an artificial intelligence-based learning model. The learning model automatically processes biosignals and derives blood glucose values without requiring manual configuration of multiple functional configurations, thereby reducing system complexity while maintaining measurement precision.
Solution Approach 2:
The learning model is trained to automatically process biosignals and generate blood glucose measurements autonomously. The system performs self-calibration and self-processing through the learning algorithm, eliminating the need for manual intervention in each processing stage and reducing overall system complexity.
2Reliability
If multiple functional configurations are individually set or manipulated by a designer, then biosignal processing can be optimized, but the workload and time required increase
Solution Approach 1:
The learning model is pre-trained using comprehensive training data and algorithms that capture the essential patterns of biosignal processing. This preliminary training incorporates the knowledge of multiple functional configurations (sampling, noise removal, filtering, calibration) into the model's parameters, so that no manual setup is needed during actual operation. The model automatically performs reliable processing based on its pre-learned patterns.
3Device complexity
If intermittent blood glucose measurement is used, then device simplicity is maintained, but accurate detection of hypoglycemic conditions becomes difficult
Solution Approach 1:
The patent enables continuous blood glucose monitoring through the learning model that processes biosignals in real-time as they are captured. The model continuously derives blood glucose values from the biosignal data stream, providing continuous monitoring capability without requiring complex hardware modifications. This continuous action allows for accurate detection of hypoglycemic conditions while maintaining relative device simplicity.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An embodiment may provide a blood glucose measurement method using a learning model, the method including obtaining a biosignal, generating biometric data from the biosignal, training a learning model with training biometric data, so as to output a blood glucose value, obtaining a blood glucose value corresponding to the biometric data from the biometric data via the learning model when training is completed, and providing, to a user, the blood glucose value corresponding to the biometric data.