Cardiopulmonary bypass (CPB) is integral to most cardiac surgeries, where perfusionists play the critical role of operating the heart-lung machine. However, errors and adverse events during CPB may occur due to stress, task complexity, information overload, or miscommunication with surgeons, anesthesiologists, and nurses. To mitigate these risks, we propose an AI-driven clinical decision support system (CDSS) to aid perfusionists in making timely and safety-critical decisions during CPB.

Building on our prior work, we developed a prototype CDSS that leverages machine learning (ML) to recommend actions consistent with perfusionist decisions in complex clinical scenarios. The system is trained on a retrospective dataset capturing real-world decision-making by expert perfusionists. The key contributions of this work include: (1) expanding the dataset of perfusionist decisions in the operating room from three to 22 patients and (2) implementing and validating two distinct ML frameworks (supervised learning and reinforcement learning) for predicting perfusionist actions.