Masters Theses

Orcid ID

0009-0005-8304-2365

Date of Award

12-2025

Degree Type

Thesis

Degree Name

Master of Science

Major

Computer Science

Major Professor

Hairong Qi

Committee Members

Xiaopeng Zhao, Audris Mockus, Hairong Qi

Abstract

Dementia poses a significant global health challenge, yet existing technology-assisted care systems often fail to adapt to the complex and evolving behaviors of patients, leading to inefficient monitoring and increased caregiver burden. This thesis addresses the critical shortcomings of current patient monitoring systems, which typically rely on static or random search strategies when patients wander, resulting in dangerous delays in detection and compromising patient safety.

To overcome these limitations, this work presents a novel intelligent multi-agent system that integrates adaptive spatial-temporal learning, computer vision, loop prevention mechanisms, and comprehensive biometric simulation within a realistic behavioral framework. The system employs an algorithm that dynamically learns individual patient movement patterns to build a probabilistic model of their location, which in turn guides an intelligent system for proactive and efficient care. A key innovation is the integration of a synthetic biometric monitoring system that generates realistic vital signs, including heart rate, HRV, and SpO2 to model stress responses, simulate medical emergencies, and objectively measure the physiological impact of care interventions. This allows for the data-driven adaptation of care strategies, which are tested and refined within a sophisticated multi-agent simulation featuring patient agents with realistic cognitive decline and caregiver agents employing evidence-based intervention techniques. The system incorporates an adaptive loop prevention framework that detects and breaks repetitive interaction patterns, utilizing confidence scoring to prioritize effective interventions while avoiding counterproductive cycles.

Through extensive simulation-based experiments, the system demonstrated a 40--60\% reduction in average patient detection time compared to baseline methods. The biometric-guided approach increased care intervention success rates from a baseline of 38\% to 55\% and achieved 94\% sensitivity in detecting simulated medical emergencies. The loop prevention system successfully identified and mitigated repetitive interaction patterns, preventing escalation of patient agitation. This research contributes a comprehensive, adaptive, and empirically validated framework that enhances patient safety, reduces caregiver burden, and shifts the paradigm from reactive monitoring to proactive, personalized, and physiologically-informed dementia care.

Download
Files over 3MB may be slow to open. For best results, right-click and select "save as..."

Share

COinS