UCLA

At a time when computational power and big data are driving revolutionary changes across various sectors, the healthcare industry is on the verge of a significant transformation. The integration of sophisticated computational techniques promises not only to enhance medical decision-making but also to fundamentally change the delivery of healthcare services. However, the sector grapples with challenges like the underutilization of its abundant data in clinical guidelines, which tend to rely on oversimplified, population-based methods, and the scarcity of annotated and labeled datasets in medical contexts. In this dissertation, we address the challenges impeding the full exploitation of computational capabilities in healthcare. The initial chapters are dedicated to enhancing decision-making at an individual level. Specifically, Chapter One addresses the classification challenges in 3D medical imaging, a task hindered by sparse and labor-intensive annotation processes. Chapter Two introduces a novel approach that leverages transformer models to augment and personalize clinical practice guidelines, thereby enhancing their relevance and applicability to individual patient care. Subsequent chapters pivot to a population-level perspective, presenting computational techniques that analyze varied datasets, ranging from social media data to records of the COVID-19 pandemic. These methods attempt to identify causal mechanisms and quantify uncertainty to support decision-making that is both data-driven and reliable.