Adaptive Regularization Networks for Irregular Glucose Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current glucose monitoring systems for diabetes management rely on time-series or linear extrapolation methods, which require frequent and consistent data input, making them inconvenient for users and less effective in predicting future glucose concentrations, especially with irregular sampling rates.
Innovation Solution
A computer-implemented method using adaptive regularization networks that predicts future glycaemic states based on irregularly sampled data, employing two learning machines to extrapolate glucose values as a continuous function of time, incorporating physiological and therapeutic data, and allowing for improved prediction accuracy with reduced data frequency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If time-series or linear extrapolation methods are used for glucose prediction, then prediction capability is provided, but frequent and consistent data input is required which reduces user convenience
Solution Approach 1:
The patent transforms the prediction approach by changing from traditional time-series parameter fitting to a regularization-based functional space method. This allows the system to work effectively with irregularly sampled data without requiring frequent consistent measurements, thereby improving user convenience while maintaining prediction reliability through adaptive regularization parameters and kernel functions.
Solution Approach 2:
The invention introduces dynamic adaptability by allowing the regularization network to automatically adjust to varying data sampling rates and patterns. The system dynamically selects appropriate kernel functions and regularization parameters based on the actual data characteristics, enabling reliable predictions whether data is sampled frequently or irregularly, thus resolving the contradiction between prediction reliability and ease of operation.
2Measurement precision
If traditional glucose monitoring systems are used, then continuous measurements can be obtained, but the systems require frequent measurements which increase device complexity and user burden
Solution Approach 1:
The patent applies the principle of partial action by demonstrating that accurate glucose prediction can be achieved with less than the traditional frequency of measurements. The regularization network can produce reliable predictions from sparsely sampled data, eliminating the need for continuous or near-continuous measurements, thereby reducing device complexity and user burden while maintaining adequate monitoring precision.
Solution Approach 2:
The invention extracts the essential predictive information from irregularly sampled glucose data using regularization networks, separating the core prediction function from the requirement for frequent measurements. This extraction allows the system to maintain measurement precision by focusing on key data points rather than requiring continuous data streams, thus reducing overall system complexity.
3Measurement precision
If regularization networks with adaptively chosen kernels are used, then prediction accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent applies preliminary action by pre-selecting from a library of kernel functions and pre-configuring regularization parameters based on expected data characteristics. This preliminary preparation allows the adaptive regularization network to achieve high prediction accuracy without performing complex real-time computations, thereby reducing computational complexity during actual glucose prediction while maintaining improved prediction accuracy.
Data Source
AI summary
The invention relates to a method and a device for predicting a glycaemic profile of a subject. A multistage algorithm is employed comprising a prediction setting stage specifying a functional space for the prediction and a prediction execution stage specifying a predicted future glycaemic state of the subject in the functional space as a continuous function of time.