Ensemble of Partitioned Sensor Glucose Models for Accuracy

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing glucose monitoring systems face challenges in accurately estimating blood glucose levels due to complex interactions between glucose sensor properties and measurement results, leading to suboptimal accuracy and large variance in measurement results.

Innovation Solution

The use of an ensemble of partitioned sensor glucose models, where the input parameter space is divided into regions based on different partition schemes, allows for the selection of regional models that better characterize the relationship between glucose levels and sensor measurement data within specific regions, thereby improving estimation accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a single glucose sensor model is used for all conditions, then the system is simple and easy to operate, but the accuracy and consistency of glucose level estimates deteriorate due to complex interactions between sensor properties and measurement results

Engineering Contradiction:
Improveaccuracy of glucose level estimatesVSAvoidmodel complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent divides the input parameter space into multiple regions based on different partition schemes (first partition scheme and second partition scheme). Each region is associated with a regional SG model that is specifically trained to characterize the relationship between glucose levels and sensor measurement data for that particular region. This segmentation allows the system to use simpler, region-specific models rather than one complex universal model, thereby improving accuracy while maintaining manageable complexity.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies local quality by creating regional SG models that are optimized for specific regions of the input parameter space. Each regional model captures the unique characteristics and relationships specific to its region, allowing for more accurate glucose estimation under different sensor conditions. This local optimization improves measurement precision without requiring the entire system to be overly complex.

Inventive Principle:
Principle #3Local quality

2Reliability

If multiple partition schemes are used to divide the input parameter space, then the accuracy and consistency of glucose level estimates improve by better characterizing regional relationships, but the device complexity increases

Engineering Contradiction:
Improveconsistency of glucose level estimatesVSAvoidmodel selection and combination complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent segments the input parameter space using multiple different partition schemes, creating a set of regions where each region is characterized by specific partitioning criteria. This segmentation enables the system to capture diverse relationships between sensor data and glucose levels across different operating conditions, improving reliability and consistency of estimates.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent combines multiple regional SG models from different partition schemes to create an ensemble that improves overall estimation reliability. By merging the strengths of multiple partitioning approaches, the system achieves more consistent glucose level estimates across varying sensor conditions while managing complexity through systematic model combination.

Inventive Principle:
Principle #5Merging (Combining)

Data Source

PatentEP4545002A1Ensemble of partitioned sensor glucose models
Publication Date: 2025.04.30 MEDTRONIC MINIMED INC
  • EP4545002A1 patent drawingFigure 1
  • EP4545002A1 patent drawingFigure 2
  • EP4545002A1 patent drawingFigure 3

AI summary

A processor-implemented method includes receiving sensor measurement data from a glucose sensor; selecting, based on the sensor measurement data, a first regional sensor glucose (SG) model from a first plurality of regional SG models for respective regions of a first plurality of regions of an input parameter space associated with the sensor measurement data, and a second regional SG model from a second plurality of regional SG models for respective regions of a second plurality of regions of the input parameter space; estimating a first SG value and a second SG value using the first regional SG model and the second regional SG model, respectively; and determining a predicted SG value based on a combination of the first SG value and the second SG value. The input parameter space is partitioned into the first plurality of regions and the second plurality of regions using different partition schemes.