Date of Award
5-1-2026
Document Type
Thesis
Degree Name
M.S. in Engineering Science
First Advisor
Kasem Khalil
Second Advisor
Md Sakib Hasan
Third Advisor
Rabiul Islam
School
University of Mississippi
Relational Format
dissertation/thesis
Abstract
Artificial Intelligence is transforming healthcare delivery, particularly in chronic disease management and patient monitoring. However, significant challenges remain in ensuring data privacy, system security, and reliable real-time monitoring while maintaining diagnostic accuracy. This thesis presents a comprehensive investigation into secure, decentralized, and intelligent healthcare systems through three interconnected research contributions addressing these critical challenges.
First, we conduct a comprehensive survey of AI applications across major healthcare domains including Chronic Kidney Disease, Diabetes Mellitus, Cardiovascular Disease, and cancer. Through systematic review of over 150 studies published between 2017 and 2024, we examine how machine learning techniques are applied to diverse medical data sources including Electronic Health Records, medical imaging such as MRI, CT, and X-ray scans, laboratory tests, and wearable sensor signals. Our analysis reveals that advanced learning methods consistently achieve predictive accuracies above 95% for disease diagnosis and patient monitoring. However, the survey identifies persistent challenges in data privacy compliance with regulations such as HIPAA and GDPR, algorithmic bias affecting equitable healthcare delivery, and barriers to integration with existing clinical workflows that limit real-world deployment.
Second, recognizing the critical importance of reliable data transmission in healthcare monitoring, we investigate how communication protocols affect the performance of Continuous Glucose Monitoring systems integrated with machine learning for diabetes management. Using network simulations with eight glucose sensor nodes, we evaluate three wireless communication approaches in terms of data transmission speed, reliability, and energy efficiency. Our findings demonstrate that protocol selection directly impacts diagnostic accuracy, with optimal configurations achieving 98% accuracy in detecting abnormal glucose levels, significantly outperforming alternative approaches that achieve 92% and 96% accuracy. These results provide practical guidance for implementing wearable monitoring devices that balance data reliability with battery life constraints.
Third, addressing the privacy and security challenges identified in our survey, we develop a secure collaborative learning framework for diabetes data management that enables multiple healthcare institutions to jointly train predictive models without sharing sensitive patient data. The system combines distributed learning with privacy-preserving mechanisms and secure verification processes to protect patient information while maintaining model quality. Experimental validation on three real-world diabetes datasets totaling over 355,000 patients demonstrates that our approach achieves diagnostic performance within 1.5% of traditional centralized methods while reducing data transmission requirements by 32% and providing protection against malicious participants and privacy attacks.
Collectively, these contributions demonstrate that integrating secure distributed learning, optimized wireless communication, and advanced machine learning techniques can significantly enhance diagnostic precision and enable continuous patient monitoring while maintaining strict data protection standards required for clinical applications. The practical implementations and comprehensive validation across diverse datasets establish feasibility for deployment in real-world healthcare environments serving millions of patients globally.
Recommended Citation
Eda, Jerina, "Federated Learning and Blockchain-Enabled Framework for Secure and Intelligent Healthcare Systems" (2026). Electronic Theses and Dissertations. 8812.
https://egrove.olemiss.edu/etd/8812