Layered Machine Learning Models for Continuous Glucose Monitoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional continuous glucose monitoring (CGM) systems produce unreliable glucose concentration estimates due to reliance on single sensor models, leading to intermittent failures and excessive data blanking, which can be fatal in cases of hypoglycemia or hyperglycemia, especially under outlier conditions such as physical activity or environmental changes.
Innovation Solution
The implementation of layered machine learning models that prioritize sensor data when available and probabilistic information when data is limited, along with micro models trained for specific outlier conditions, to generate more reliable glucose values and reduce blanking.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single sensor glucose estimation model is used, then the device complexity is reduced, but the reliability of glucose monitoring deteriorates under outlier conditions
Solution Approach 1:
The patent divides the single sensor glucose estimation model into multiple specialized models (e.g., first model for normal conditions, second model for outlier conditions). Each model is trained on specific datasets corresponding to different operational conditions, allowing the system to segment the problem space and improve reliability without requiring a single overly complex model.
Solution Approach 2:
The system dynamically selects which model to use based on real-time sensor data characteristics. When outlier conditions are detected (e.g., physical activity, environmental changes), the system switches to the appropriate specialized model. This dynamic adaptation allows the system to maintain high reliability across varying conditions without permanently increasing device complexity.
2Device complexity
If conventional CGM systems use single model estimation, then the system operates simply, but excessive data blanking occurs leading to loss of glucose information
Solution Approach 1:
By segmenting the estimation task across multiple condition-specific models, the system can provide continuous glucose estimates even when sensor data quality varies. This prevents the need to blank data during outlier conditions, as specialized models can reliably estimate glucose values under those conditions.
Solution Approach 2:
The patent introduces probabilistic information as an intermediary element that bridges sensor data and glucose estimates. When sensor data is limited or unreliable, the system uses probabilistic models to generate pseudo-future estimates, maintaining continuous information flow without excessive blanking.
3Reliability
If probabilistic information is prioritized when sensor data is limited, then glucose estimation reliability improves, but device complexity increases
Solution Approach 1:
The patent segments the estimation approach into distinct layers: sensor data-driven models for normal conditions and probabilistic models for limited data conditions. This segmentation allows the system to use complex probabilistic reasoning only when necessary, rather than always, thereby managing device complexity while improving reliability under data-limited conditions.
4Reliability
If multiple layered models are implemented, then compliance with regulatory standards improves, but the device complexity increases
Solution Approach 1:
By training separate models on different datasets (e.g., one model trained on data meeting iCGM criteria, another on outlier conditions), the system can demonstrate compliance with regulatory standards across various operating conditions. This segmented approach provides the documentation and performance evidence required for regulatory approval.
Solution Approach 2:
The layered model architecture serves multiple functions: it provides compliant glucose estimates under normal conditions, maintains reliability during outlier conditions, and generates probabilistic information for regulatory validation. This multi-functionality justifies the increased device complexity by delivering comprehensive performance across all required operational scenarios.
Data Source
AI summary
Methods, systems, and devices for improving continuous glucose monitoring (“CGM”) are described herein. More particularly, the methods, systems, and devices describe applying micro machine learning models to generate predicted sensor glucose values. The system may use the predicted sensor glucose values to display a sensor glucose value to a user. The layered models may generate more reliable sensor glucose predictions across many scenarios, leading to a reduction of sensor glucose signal blanking. The methods, systems, and devices described herein further comprise applying a plurality of micro model to estimate sensor glucose values under outlier conditions. The system may prioritize the models that are trained for certain outlier conditions when the system detects those outlier condition based on the sensor data.