Iris Feature Extraction for Non-Invasive Blood Glucose Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for blood glucose monitoring in diabetes management, such as finger-prick tests and continuous glucose monitors, are invasive, costly, inconvenient, and pose significant challenges for consistent monitoring due to discomfort, high expenses, and complexity, particularly for low-income patients and those on the go.
Innovation Solution
A non-invasive system using a smartphone-based ophthalmic imaging method that employs a multi-stage deep transfer learning computer vision model to predict blood glucose levels by capturing eye images, training convolutional neural networks to classify and extract iris feature vectors, and applying extreme gradient boosting for regression to provide accurate glucose level predictions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If finger-prick tests or continuous glucose monitors are used, then blood glucose monitoring accuracy is improved, but patient comfort and ease of operation deteriorate due to invasive procedures
Solution Approach 1:
The patent replaces the mechanical invasive system (lancet pricking, blood extraction, test strip insertion) with an optical system that captures images of the eye's iris and uses machine learning to predict blood glucose levels. This substitution eliminates physical invasion while maintaining monitoring capability through a different measurement modality (optical imaging instead of biochemical analysis of blood).
Solution Approach 2:
The patent introduces an intermediary approach by using the eye's iris as a proxy indicator for blood glucose levels. Instead of directly measuring glucose in blood or interstitial fluid, the system uses optical characteristics of the iris that correlate with glucose levels, mediated through machine learning algorithms that establish the relationship between iris imaging data and actual blood glucose values.
2Duration of action of stationary object
If continuous glucose monitors are used, then continuous monitoring capability is improved, but cost and device complexity increase significantly
Solution Approach 1:
The patent makes the smartphone camera multi-functional by using it for both standard photography and blood glucose monitoring. The existing camera hardware is repurposed to capture iris images, eliminating the need for specialized expensive sensors. The machine learning model processes these images to extract glucose-related information, allowing a common device to serve multiple purposes including health monitoring.
Solution Approach 2:
The patent uses a digital copy approach by capturing images of the iris and processing them through machine learning models to infer blood glucose levels. This creates an indirect digital representation that correlates with physiological state, avoiding the need for physical sensors implanted in the body while achieving continuous monitoring through repeated imaging.
3Measurement precision
If finger-prick tests are used, then blood glucose measurement is obtained, but time consumption and operational steps increase
Solution Approach 1:
The patent performs preliminary action by pre-training machine learning models on large datasets of iris images paired with corresponding blood glucose measurements. This pre-training establishes the correlation patterns between iris characteristics and glucose levels in advance, so that during actual use, the system can quickly process new images and provide immediate predictions without requiring complex real-time analysis or multiple operational steps.
Data Source
AI summary
A method of predicting a blood glucose level of a user, comprising: obtaining, by an image capturing device, an iris image of the user; training a first convolutional neural network of a computing device using the iris image as an input to obtain a classification of the iris image; training a second convolutional neural network of the computing device using the classification and the iris image to extract an iris feature vector; and predicting, by the computing device, the blood glucose level of the user based on the iris feature vector.