The autonomic nervous system (ANS) plays a vital role in regulating essential bodily functions, including heart rate, blood pressure, respiratory rate, and emotional responses. Affective states such as pain and stress, which are closely linked to the ANS, are critical to managing a wide range of health conditions. As healthcare evolves toward a more personalized, precise, and preventive model, the integration of biosignal monitoring—enabled by advanced wearable devices—and machine learning for the accurate assessment and intervention of affective states is becoming increasingly vital. These technologies provide real-time insights into physiological responses, enabling tailored and timely interventions that significantly enhance patient outcomes, embodying the principles of next-generation healthcare.
This dissertation presents a comprehensive framework for the precision assessment and intervention of affective states, specifically focusing on pain, which is closely linked to the autonomic nervous system (ANS). Leveraging the rapid advancement of wearable devices and machine learning techniques, this research bridges the gap between subjective experience and objective measurement in healthcare. The study is structured into three main components: validation of wearable devices, development of pain assessment models, and the design of a recommendation system for smart pain management.
First, the accuracy of wearable devices such as the Oura Ring and Samsung SmartWatch was rigorously validated against medical-grade electrocardiography (ECG) monitors, ensuring reliable data collection for biosignal monitoring. This validation serves as a foundation for the subsequent contributions of the thesis. The second major contribution lies in developing robust models for pain assessment, utilizing both unimodal and multimodal machine learning approaches. These models significantly improve the accuracy and reliability of detecting pain and stress, providing a more objective basis for clinical evaluation.
A critical innovation is the derivation of additional biosignals from existing data, which enhances the predictive power of the models. By extracting nuanced physiological features that are often overlooked, this work contributes to a deeper understanding of the physiological underpinnings of pain.
The research culminates in the creation of a smart recommendation system for pain management, specifically tailored for optimizing morphine dosing in critical care settings. This system integrates biosignal data with advanced reinforcement learning algorithms, including Conservative Q-learning (CQL), to provide personalized treatment recommendations that enhance clinical decision-making. This framework addresses the challenges of pain and stress management and sets the stage for next-generation healthcare characterized by precision, personalization, and proactive care.
The organization of this thesis is as follows: Chapter 1 introduces the research problem, outlines the motivations and objectives, and reviews the relevant literature. Chapter 2 details the validation process for wearable devices, focusing on biosignal monitoring accuracy. Chapter 3 explores the derivation of additional biosignals from existing data to enhance pain assessment accuracy. Chapter 4 delves into pain assessment methodologies and presents the key findings. Chapter 5 discusses the design and implementation of the smart recommendation system for pain management. Finally, Chapter 6 concludes with a summary of the key findings, implications for future research, and potential clinical applications of the proposed framework in precision healthcare.
