Nonparametric Glucose Prediction for Patient-Specific Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing glucose-insulin models, both white-box and black-box, struggle to accurately predict blood glucose levels due to inter- and intra-patient variability, leading to inefficiencies in model-based glucose prediction and control systems for type-1 diabetes management.
Innovation Solution
Employing a patient-specific, individualized linear black-box model using a non-parametric approach with a Stable-Spline Kernel and Kalman filter to predict future blood glucose concentrations, addressing collinearity issues through input transformation and optimizing kernel hyperparameters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If parametric model-learning algorithms are used to predict blood glucose concentrations, then the model structure is simpler and easier to implement, but the prediction accuracy is insufficient due to inter- and intra-patient variability
Solution Approach 1:
The patent transitions from parametric models with fixed structural parameters to non-parametric models where the impulse response functions are estimated without assuming a specific parametric form. This allows the model to adapt to individual patient characteristics and temporal variations, significantly improving prediction accuracy while maintaining computational tractability through kernel-based methods in RKHS.
Solution Approach 2:
The patent employs time-varying impulse response functions that can adapt to intra-patient variability over time. The non-parametric approach allows the model dynamics to change according to individual patient responses rather than being constrained by fixed parametric structures, enabling the system to capture evolving physiological patterns.
2Measurement precision
If individualized patient-specific models are created to address inter- and intra-patient variability, then the prediction accuracy improves, but the model complexity and computational requirements increase
Solution Approach 1:
The patent replaces complex parametric model identification procedures with non-parametric methods based on Gaussian regression and kernel functions in RKHS. This substitution simplifies the mathematical framework by avoiding the need to specify and estimate multiple parametric parameters, while still achieving individualized patient-specific modeling through data-driven impulse response estimation.
Solution Approach 2:
The patent develops a universal non-parametric framework that can be applied to all patients without requiring patient-specific parameter tuning. The kernel-based approach in RKHS provides a unified methodology that automatically adapts to individual patient characteristics through the data, eliminating the need for separate model development procedures for each patient.
3Measurement precision
If non-parametric approach with RKHS is used to estimate impulse response functions, then the prediction accuracy significantly improves, but the computational complexity increases
Solution Approach 1:
The patent transforms the infinite-dimensional non-parametric estimation problem in RKHS into a finite-dimensional optimization problem by exploiting the reproducing kernel property. This allows the impulse response functions to be expressed as linear combinations of kernel functions evaluated at training data points, reducing computational complexity while maintaining the flexibility of non-parametric modeling.
Data Source
Figure 1~2B
Figure 3~4
Figure 5A~5B
AI summary
A method of predicting future blood glucose concentrations of an individual patient includes: identifying an individualized linear black box model of glucose-insulin by estimating a plurality of impulse response functions each accounting for an input-output relation of a plurality of individualized patient data sets, the impulse response functions being functions in a Reproducing Kernel Hilbert Space (RKHS); and applying a linear predicting technique to the selected model using the identified impulse response functions to obtain a predicted blood glucose concentration of the individual patient at a future time.