Revolutionizing healthcare with federated reinforcement learning: from machine learning to machine unlearning

Revolutionizing healthcare with federated reinforcement learning: from machine learning to machine unlearning

The landscape of healthcare is undergoing a transformative shift with the emergence of artificial intelligence (AI) and machine learning (ML) technologies, particularly in remote patient monitoring systems. These systems offer real-time data on patients’ health conditions, enabling healthcare professionals to make informed decisions and improve patient outcomes. This doctoral thesis presents a comprehensive investigation into the role of AI in enhancing patient monitoring systems, focusing on innovations in federated learning, reinforcement learning, and machine unlearning across various healthcare settings including remote patient monitoring, personalized activity tracking, mental health facilities, and predictive monitoring. The research outcomes reveal significant advancements in remote patient monitoring through AIpowered systems, enabling early anomaly detection and personalized care. FedStack, a novel federated learning architecture designed for personalized activity monitoring in remote patient monitoring systems, is introduced. Experimental results demonstrate its effectiveness in surpassing traditional approaches and optimizing sensor placement for activity recognition. Furthermore, multi-agent deep reinforcement learning models empower healthcare professionals to predict patient behaviors and take proactive interventions. The exploration of multimodality fusion and graph-enabled techniques demonstrates the potential of comprehensive smart healthcare systems that integrate diverse data sources and enable informed decision-making. Additionally, the thesis introduces a Graph-enabled Reinforcement Learning framework for time series forecasting, leveraging graphical neural networks to outperform traditional models in dynamic environments. The thesis also explores the emerging field of machine unlearning, investigating techniques to address privacy and security concerns. Explainable AI frameworks, such as QXAI, contribute to the reliability and interpretability of patient monitoring systems, fostering trust and collaboration between AI and human experts. FRAMU, a federated reinforcement learning framework with attention-based machine unlearning, is introduced, demonstrating its effectiveness in improving model performance and preserving privacy in dynamic data environments. QXAI, an explainable AI framework for quantitative analysis in patient monitoring, achieves state-of-the-art results in heart rate prediction and activity classification, enhancing model interpretability. While the research demonstrates promising outcomes, it acknowledges certain limitations, including data scale, explainability, and data privacy concerns. Future directions such as dynamic clustering, predictive vital sign monitoring, ensemble methods, and continued progress in machine unlearning are proposed to address these limitations and propel AI-driven patient monitoring systems further. This thesis makes significant contributions to the domain of AI-driven patient monitoring systems, paving the way for personalized, proactive, and effective healthcare delivery globally. It posits that the transformative potential of AI in healthcare is within reach, with continued research and innovation shaping the future of patient care, where AI-driven monitoring becomes an indispensable tool in enhancing patient well-being and transforming healthcare practices.

Files associated with this item cannot be displayed due to copyright restrictions.