Differential-Cell Deduction Learning for Non-Invasive Glucose Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current non-invasive blood glucose monitoring systems face challenges in achieving personalized and precise predictions due to large variability in human physiology and the lack of strong correlation between photoplethysmography (PPG) optically-derived features with blood glucose levels, with conventional induction learning methods being impractical for daily use due to the discomfort of frequent data collection through finger-pricks.
Innovation Solution
A deduction learning model utilizing a differential cell that calculates predicted blood glucose levels based on the correlation between differences in PPG signals and reference glucose levels, allowing for accurate predictions with minimal data collection, enhanced by a screening module to filter outliers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional induction learning methods are used for non-invasive blood glucose monitoring, then the system can provide blood glucose estimations, but the prediction accuracy is insufficient due to large variability in human physiology and weak correlation between PPG features and blood glucose levels
Solution Approach 1:
The patent introduces an intermediary calibration process using a small number of finger-prick measurements to establish personalized mapping relationships between PPG features and blood glucose levels. This intermediary step bridges the weak correlation gap by creating subject-specific calibration curves that translate optical PPG measurements into accurate glucose estimates without requiring continuous invasive sampling.
Solution Approach 2:
The system dynamically adjusts prediction parameters by continuously updating the relationship between PPG signal characteristics and blood glucose levels based on periodic reference measurements. The model adapts parameters such as correlation coefficients and transformation factors to account for physiological variations, thereby improving prediction accuracy over time while maintaining reliability across different individuals.
2Measurement precision
If frequent finger-prick data collection is performed for model training, then personalized prediction accuracy can be improved, but user comfort and practicality deteriorate due to pain and discomfort
Solution Approach 1:
The patent applies partial action by requiring only a minimal number of finger-prick measurements (e.g., 3-5 readings) during an initial calibration phase rather than frequent continuous sampling. This partial data collection is sufficient to establish personalized prediction models, achieving high accuracy while minimizing user discomfort. The system then uses these calibrated models for extended periods without requiring additional invasive measurements.
Solution Approach 2:
The system performs preliminary calibration using a small set of reference blood glucose measurements before deploying the personalized prediction model for routine non-invasive monitoring. This preliminary action establishes the individualized mapping relationships in advance, allowing the system to operate accurately for weeks or months without requiring users to undergo painful finger-pricks again, thereby greatly improving ease of operation.
3Measurement precision
If personalized modeling is implemented to account for individual physiological differences, then prediction accuracy for specific subjects improves, but system complexity increases due to the need for subject-specific calibration
Solution Approach 1:
The patent implements personalized modeling by adjusting a limited set of calibration parameters rather than creating entirely separate complex models for each subject. The system uses a unified PPG analysis framework with adaptable parameters such as scaling factors, offset values, and correlation coefficients that are customized through simple calibration procedures. This approach achieves subject-specific accuracy while keeping the overall system architecture relatively simple and manageable.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The deduction learning model significantly improves prediction accuracy and reduces the need for frequent invasive data collection, achieving high precision and reliability in non-invasive blood glucose monitoring.
Implementation Method 1
photoplethysmography (PPG), an optical signal measurement technique based on near-infrared (NIR) transmittance or reflectance
Implementation Method 2
photoplethysmography (PPG), an optical signal measurement technique based on near-infrared (NIR) transmittance or reflectance
Data Source
AI summary
The present invention relates to a system and method for predicting blood glucose levels in a non-invasive manner using deduction learning. The deduction learning (DL) model 10 of the present invention predicts blood glucose level 300 based on the relationship or correlation between blood glucose level variation and PPG signal variation. Specifically, the DL model 10 of the present invention comprises a differential cell 100 (DC) configured to calculate predicted blood glucose level BGpred 300 using two PPG signals Si 200b and Si−1 200a, a BG i−1,ref 230 and relationship or correlation between variances of the two PPG signals and variances of the reference blood glucose level 230 and the predicted blood glucose level 300.