Bridging Human and Artificial Intelligence : Machine Learning, Data Platforms, and Decision Support Systems in Sleep Research

Abstract

Artificial Intelligence (AI) delivers groundbreaking automation capabilities to tasks that historically require manual human labor. However, its integration into fields like healthcare remains challenging due to concerns around interpretability, data standardization, and clinical trust. This thesis comprehensively explores AI's potential to enhance sleep medicine by addressing these challenges. This work offers a holistic perspective on AI in sleep research, spanning the journey of data from collection and augmentation to its final presentation to human experts as well as the lifecycle of AI, from its inception to its integration into sleep medicine workflows. The key findings include a novel respiratory cycle detection algorithm with 94\% accuracy, insights into clustering respiratory events via unsupervised learning, and evidence that AI-assisted workflows reduce scoring time by up to 65 minutes while improving inter-rater agreement among sleep technologists. Furthermore, our research confirms that sleep technologists can work effectively alongside AI without significant distrust, highlighting a high level of clinical acquiescence. The contributions focus on three key areas: (1) developing algorithms rooted in physiological principles to improve interpretability, (2) creating standardized data pipelines for scalable and reproducible AI deployment and (3) integrating human-in-the-loop solutions to enhance clinical decision-making. These advancements underscore the transformative potential of AI in sleep medicine, providing a holistic view of its integration into clinical workflows. This research paves the way for the broader adoption of AI in healthcare by fostering trust, efficiency, and interpretability.

Artificial Intelligence (AI) delivers groundbreaking automation capabilities to tasks that historically require manual human labor. However, its integration into fields like healthcare remains challenging due to concerns around interpretability, data standardization, and clinical trust. This thesis comprehensively explores AI's potential to enhance sleep medicine by addressing these challenges. This work offers a holistic perspective on AI in sleep research, spanning the journey of data from collection and augmentation to its final presentation to human experts as well as the lifecycle of AI, from its inception to its integration into sleep medicine workflows. The key findings include a novel respiratory cycle detection algorithm with 94\% accuracy, insights into clustering respiratory events via unsupervised learning, and evidence that AI-assisted workflows reduce scoring time by up to 65 minutes while improving inter-rater agreement among sleep technologists. Furthermore, our research confirms that sleep technologists can work effectively alongside AI without significant distrust, highlighting a high level of clinical acquiescence. The contributions focus on three key areas: (1) developing algorithms rooted in physiological principles to improve interpretability, (2) creating standardized data pipelines for scalable and reproducible AI deployment and (3) integrating human-in-the-loop solutions to enhance clinical decision-making. These advancements underscore the transformative potential of AI in sleep medicine, providing a holistic view of its integration into clinical workflows. This research paves the way for the broader adoption of AI in healthcare by fostering trust, efficiency, and interpretability.