PREDICTION AND CLOSED-LOOP CONTROL OF BLOOD PRESSURE FOR HEMORRHAGE RESUSCITATION

Abstract

Hemorrhage is responsible for a large percentage of mortality worldwide and the majority of fatalities on the battlefield. Resuscitation procedures for hemorrhage trauma patients are critical for their recovery. Currently, during resuscitation, physicians manually monitor blood pressure and use intuition to determine when fluid should be administered and how much. Due to factors such as exhaustion, distraction, and inexperience of the physician, this method has often been reported as fallible. This thesis proposes two methods to assist in automating hemorrhage resuscitation. The first is a blood pressure prediction algorithm for decision support systems. The algorithm individualizes itself to different subjects using extended Kalman filtering (EKF), to account for high inter-subject variability, before accurately forecasting future blood pressure. The second method is an observer-based feedback controller which regulates blood pressure from a hypotensive state back to a “healthy” setpoint. The controller was designed using linear matrix inequality (LMI) techniques to ensure it was absolutely stable, which let a portion of the hemodynamic plant model remain unspecified and allowed for performance over a range of physiologies. Both strategies were evaluated in-silico on a cohort of 100 virtual patients generated from an experimental dataset. The prediction algorithm showed accuracy superior to conventional assumptions. The controller tracked the given setpoint with an accuracy and performance comparable to more complex adaptive methods. Further work, with respect to the prediction algorithm, includes developing it into a full decision-support system and incorporating disturbance rejecting components to account for common issues such as rebleed. The controller’s performance deteriorates for high-speed applications, suggesting further study is required to increase its situational flexibility.