Shelby Hall Graduate Research Forum Posters

Low-Cost Wearable Edge-AI Device for Diabetes Management

Low-Cost Wearable Edge-AI Device for Diabetes Management

Description

Diabetes Mellitus (DM) refers to a set of serious metabolic disorders affecting 537 million people worldwide, according to 2021 statistics from the International Diabetes Federation. Blood glucose (BG) levels are a vital health indicator for managing DM, but conventional methods of measuring BG use direct blood or interstitial fluid samples. While accurate, these ‘invasive’ methods can cause discomfort, tissue damage, pose a risk of infection for the patient, and often require expensive consumable materials or devices to use. Developing a ‘noninvasive’ method of measuring or approximating BG has grown increasingly relevant as healthcare technology advances and cases of DM continue to rise. Artificial intelligence (AI) techniques have shown promising results in BG prediction using a variety of noninvasively sampled signals. However, these techniques are seldom implemented on wearable edge devices owing to complex sensors, costly algorithms, and reliance on cloud computing. This work designs, prototypes, and evaluates a standalone wearable device to estimate BG with AI methods, with emphasis placed on integration of software, hardware, and user interaction. Static information from user input and a photoplethysmography sensor are used to provide noninvasive data for the AI model of this device to estimate BG. The device achieves a prediction accuracy of 16.8% mean average percentage error and produces mostly clinically acceptable results according to Clarke Error Grid Analysis.

Publication Date

3-2025

Department

Electrical & Computer Engineering

City

Mobile

Disciplines

Biomedical | Electrical and Electronics | Endocrinology, Diabetes, and Metabolism | Other Electrical and Computer Engineering

Low-Cost Wearable Edge-AI Device for Diabetes Management

Share

COinS