AI-ENABLED BIOMEDICAL SYSTEMS: FROM PHYSIOLOGICAL RISK STRATIFICATION TO AUTONOMOUS BIOPRINTING OPTIMIZATION

Abstract

Advancements in machine learning have enabled automated solutions across a diverse array of healthcare-related tasks, including waveform analysis, image-based modeling, and adaptive control. Integration of these tools into traditional clinical and bioengineering workflows has the potential to enhance decision-making in complex, high-dimensional environments. In both critical care medicine and biofabrication, key tasks—such as prognostication of adverse events and optimization of print quality—require expert-driven, multi-objective reasoning that is often difficult to formalize. These challenges represent domains in which artificial intelligence can provide meaningful support. This dissertation develops and evaluates machine learning workflows across two such domains: (1) physiological signal-based modeling for clinical risk stratification and (2) semi-autonomous optimization of printability in bioprinting systems. In the clinical domain, this work investigates the utility of photoplethysmography (PPG) as a source of physiological information for outcome-relevant modeling in trauma and critical care settings. In parallel, this dissertation advances the development of AI-enabled experimental systems for tissue engineering. The framework enables structured, semi-autonomous optimization of bioprinting processes through integration of hardware control, real-time sensing, and algorithmic decision-making. A central contribution is the characterization and calibration of temperature dynamics within the bioprinting system, including discrepancies between tool-reported and bulk bioink temperature and the modeling of equilibration behavior. Collectively, this work demonstrates both the opportunities and limitations of AI-enabled biomedical systems, emphasizing the importance of integrating data-driven methods with domain-specific constraints to support clinical decision-making and experimental design.