CGM Sensor Error Detection Using Multi-Signal Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current continuous glucose monitoring (CGM) systems face challenges in scaling multi-dimensional input signals, struggle with excessive data loss due to blanking, and lack real-time error detection and correction capabilities, which can lead to inaccurate readings and potential health risks.
Innovation Solution
Implementing a machine learning model that classifies multi-dimensional CGM sensor data using iCGM criteria to identify and correct outliers and erroneous sensor use conditions, incorporating training data to enhance accuracy and compliance with regulatory standards.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If current CGM systems use single signal analysis to determine accuracy, then the system complexity is low and ease of operation is maintained, but the measurement precision and reliability are insufficient to meet multi-dimensional regulatory criteria
Solution Approach 1:
The patent transitions from single-signal analysis to multi-dimensional signal analysis by incorporating multiple sensor signals (Isig, EIS, Vcntr) and their derivatives into the accuracy determination process. This dimensional expansion enables the system to meet stricter regulatory criteria (iCGM) by considering multiple factors simultaneously, thereby improving measurement precision without being constrained by single-signal limitations.
Solution Approach 2:
The patent introduces machine learning models as intermediary components that process multi-dimensional sensor signals and regulatory criteria. These models act as mediators between the complex multi-signal input and the accuracy determination output, managing the complexity by learning patterns from training data and providing accurate predictions without requiring manual configuration of complex analysis rules.
2Reliability
If CGM systems implement comprehensive multi-dimensional input analysis, then the measurement precision and regulatory compliance improve, but the device complexity and computational requirements increase
Solution Approach 1:
The patent implements preliminary action by training machine learning models offline using comprehensive multi-dimensional sensor data and regulatory criteria before deployment. During actual CGM operation, the pre-trained models quickly process sensor signals to determine accuracy, avoiding the need for real-time complex computations. This separates the computationally intensive training phase from the operational phase, improving reliability without burdening the device with excessive real-time computational requirements.
Solution Approach 2:
The machine learning models perform self-service by automatically learning from training data the complex relationships between multi-dimensional signals and accuracy criteria. Once trained, the models independently process sensor data and determine accuracy without requiring manual intervention or complex rule-based systems, thereby improving reliability while keeping the operational device complexity manageable.
3Adaptability or versatility
If CGM systems use traditional accuracy determination methods, then the ease of manufacture and deployment is maintained, but the adaptability to new regulatory criteria and error patterns is limited
Solution Approach 1:
The patent implements dynamics by using machine learning models that can be retrained and updated with new training data reflecting changing regulatory criteria or error patterns. Unlike static rule-based systems, these dynamic models adapt to new requirements by learning from updated training sets, enabling the CGM system to maintain high adaptability to regulatory criteria while preserving ease of deployment through automated training procedures and pre-packaged model solutions.
Data Source
AI summary
Techniques for improving continuous glucose monitoring (“CGM”) are described herein. In some embodiments, the techniques involve obtaining sensor data; applying, to the sensor data, a machine learning model trained to identify sensor data error patterns; and detecting an erroneous sensor use condition based on output of the machine learning model indicating an error pattern identified in the sensor data.