Real-Time Inference and Adaptive Closed-Loop Control of Physiological States in Critical Care: A Cardiovascular Case Study

Clinicians in the operating room and intensive care unit must perform real-time inference and regulation of physiological states to properly care for their patients. For example, if arterial blood pressure (ABP) gets too low for too long, organs don’t get enough oxygen and subsequently get injured. In the face of low ABP, clinicians must select a treatment that counteracts the difficult-to-measure cause (e.g., abnormal heart function, vascular function or total blood volume). Then, they must continuously monitor and adjust their selected treatment (e.g., an intravenous medication or fluids) until their patient recovers. This is a formidable task for the following reasons: physiological systems vary not only between patients, but also within a single patient over time; access to low-risk, accurate physiological measurements is often limited; and the actuators available to regulate a physiological system are slow and imprecise. Because of these challenges, much of this clinical inference and decision-making process should be automated for more personalized, consistent and objective patient care.

In this talk, I will present my work towards automating the primary cardiovascular management goal of regulating ABP. First, I will present a real-time method for estimating cardiac and vascular function from a single ABP waveform, as validated with swine data [IEEE 2024, selected as featured article]. This method enabled the discovery that such states can be reliably estimated independently of the minimally invasive ABP catheter location*.* Second, I will present an adaptive closed-loop system for ABP control that I developed, from theory to in vivo experiments, now undergoing validation in swine. This system is patient-specific without relying on prior patient data, is informed by mechanistic cardiovascular states [ACC 2025, invited article] and has achieved error tolerance levels comparable to natural physiological variability. Third, I will present observations of the joint dynamics between ABP and general anesthesia-induced electroencephalography signals, enabled by my adaptive closed-loop control system. These captured dynamics will inform a control system regulating both ABP and unconsciousness under general anesthesia. Finally, I will outline the clear path towards translating this system to human clinical use.

 

Bio: