Edge AI and deep learning for precision decision support in diabetes diagnosis and management
No Thumbnail Available
Authors
Meeting name
Sponsors
Date
Journal Title
Format
Thesis
Subject
Abstract
Diabetes is a complex of metabolic disorders that disrupts the autonomous regulation of blood glucose concentration in human body. This chronic condition requires consistent medical monitoring and management to avert acute complications. Inadequate glycemic control can result in impaired cellular function and tissue damage across multiple organs, ultimately culminating in organ failure. Effective glucose regulation demands continuous monitoring and precise decision-making by individuals with the disease. Artificial intelligence (AI)-enabled clinical decision support system (CDSS) is a promising approach to enhance the diagnosis and management of the disease. This dissertation proposes and develops efficient edge AI and deep learning (DL) techniques to address the challenges associated with the diagnosis and management of diabetes. The first study uses medical health records implementing different machine learning (ML) methods to develop a risk assessment tool to generate user-specific binary recommendations (high and low chances of developing diabetes). The pre-trained ML model embedded into a smartphone application takes answers to patient-related questions as input and provides feedback on the likelihood of developing diabetes. The second study implements a neural network inference on a Field Programmable Gate Array (FPGA) to automatically predict diabetes from clinical measurements. The inference of the best-trained model has been realized in hardware to develop intelligent diabetes detection tools. The third study seeks to build an advisory tool utilizing predictive analytics to generate personalized blood glucose forecasts. Deep multitask learning (MTL) is proposed to enhance the personalized prediction of blood glucose levels. Consistently leading performance was obtained within the clinically acceptable regimen for multi-step prediction horizons. Finally, an ablation study comprehensively investigates the systematic integration of highly relevant life events and physiological parameters with continuous glucose monitoring (CGM) data to understand the synergy in glucoregulatory systems. The proposed edge AI and deep learning tools and techniques demonstrate promising results in diabetes risk assessment, diagnosis, and personalized management with the potential integration into the CDSS and aid in medical decisions through better therapeutic and lifestyle modification.
Table of Contents
DOI
PubMed ID
Degree
Ph. D.