ADVANCED HBA1C PREDICTION MODEL FOR PROACTIVE DIABETES MANAGEMENT

LICENSING & PARTNERING OPPURTUNITIES


This invention introduces an advanced HbA1c Prediction Model designed to revolutionize diabetes management. Developed by a team of distinguished professionals, this innovative model leverages time series CGM sensor data, convolutional neural network (CNN), and few-shot learning (FSL) techniques for accurate and proactive HbA1c level prediction. The model demonstrates an unprecedented accuracy of 92.30%, marking a significant milestone in the realm of diabetes care.

The Problem

The glycated hemoglobin (HbA1c) serves as a critical biomarker for diabetes management. While regular monitoring is essential, the absence of an advanced prediction model has hindered proactive interventions. Elevated HbA1c levels pose risks of diabetes-related complications, emphasizing the need for early predictions to facilitate timely adjustments in treatment plans and lifestyle. Traditional approaches lack the accuracy demanded for such predictions, necessitating a novel solution.

The Solution

The Advanced HbA1c Prediction Model pioneers the conversion of time series CGM sensor data into binary and histogram images. These images undergo feature extraction through a convolutional neural network (CNN) employing few-shot learning (FSL) techniques. The normalized FSL-distance (FSLD) metric is introduced for precise image differentiation based on HbA1c levels. The final prediction is accomplished using a k-nearest neighbor (KNN) model with majority voting, achieving an unprecedented accuracy of 92.30%


Inventors

Goran Petrovskicc

Sidra Medicine

Marwa Qaraqe

Hamad Bin Khalifa University

Md Shafiqul Islam

Hamad Bin Khalifa University

Samir Brahim Belhouar

Hamad Bin Khalifa University