Layered ML Models for CGM Signal Blanking Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional continuous glucose monitoring (CGM) systems intermittently produce unreliable estimates of glucose concentration, leading to potential severe or fatal effects due to neglecting or exacerbating hypoglycemia and hyperglycemia.
Innovation Solution
The implementation of a layered machine learning model system that trains multiple models to predict sensor glucose values based on both sensor data and probabilistic information, allowing the system to prioritize data types specific to given situations, such as using sensor data when available and probabilistic information when data is limited.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional CGM systems use a single sensor glucose estimation model, then the system structure is simple, but the system produces unreliable estimates and excessive blanking when sensor data is unavailable or limited
Solution Approach 1:
The patent divides the single estimation model into multiple specialized models (first model for sensor data, second model for probabilistic information, third model for hybrid input). Each model is trained for specific conditions and can be selectively applied based on data availability, improving reliability without requiring complete system redesign.
Solution Approach 2:
The system dynamically selects which model to use based on real-time sensor data quality and availability. When sensor data is reliable, the first model is used; when limited, the second model takes over; when both are available, the third model provides optimal estimates. This dynamic adaptation resolves the contradiction between reliability and complexity.
2Measurement precision
If the system uses multiple machine learning models with different data characteristics, then the accuracy of glucose measurements improves, but the computational complexity and processing time increase
Solution Approach 1:
Each model in the plurality is specialized for specific data characteristics and conditions. The first model optimizes for sensor data quality, the second for probabilistic information, and the third for combined inputs. This local specialization allows each model to achieve high precision in its domain without requiring all models to be equally complex.
Solution Approach 2:
The system applies models selectively based on data availability rather than always using all models. When sensor data is sufficient, only the first model runs; when probabilistic information is needed, the second model activates. This partial application reduces overall computational complexity while maintaining measurement precision when needed.
3Measurement precision
If the system blanks sensor glucose signals when data is unreliable, then measurement accuracy is protected, but valuable glucose data is lost and monitoring continuity is disrupted
Solution Approach 1:
The second model acts as an intermediary that generates probabilistic information to bridge gaps when sensor data is unavailable or unreliable. Instead of completely blanking the signal, the system uses the second model to produce estimated glucose values based on probabilistic patterns, maintaining monitoring continuity while protecting against inaccurate readings.
Solution Approach 2:
The system prepares multiple models in advance, each trained to handle specific data conditions. When sensor data quality degrades, the pre-trained second and third models are already ready to provide cushioning estimates, preventing complete data loss and maintaining monitoring continuity without compromising accuracy.
4Speed
If the system relies heavily on sensor data, then real-time monitoring is improved, but the system becomes ineffective when sensor data is periodically unavailable
Solution Approach 1:
The system creates a multi-functional model architecture where the first model handles real-time sensor data processing, the second model handles probabilistic information when sensor data is unavailable, and the third model combines both inputs. This universality allows the system to maintain real-time monitoring capabilities across diverse operating conditions, resolving the contradiction between speed and adaptability.
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.