Adaptive Predictive Control of Arterial Blood Pressure Based on a Neural Network During Acute Hypotension

Abstract

In acute hypotension, an automated drug infusion system to control mean arterial blood pressure (MAP) has not been previously studied, though many investigations have examined the use of vasodilating drugs to control MAP in postoperative hypertension. Therefore, we examined an automated control of MAP during acute hypotension using a neural network (NN) approach. A proportional-integral-derivative (PID) control, an adaptive predictive control using a NN (APC(NN)), a combined control of APC(NN) and PID (APC(NN-PID)), a fuzzy control, and a model predictive control were tested in computer simulation based on the MAP response to norepinephrine (NE) of 25 microg ml(-1). In six anesthetized rabbits, using the NE of 25 microg ml(-1), the PID control, APC(NN), and APC(NN-PID) prevented severe hypotension compared to an uncontrolled condition. Under PID control, four of the six animals showed MAP oscillation. Using NE of 50 microg ml(-1), the rabbits recovered from acute hypotension for all systems tested but showed sustained MAP oscillation during PID control. In conclusion, utilization of a NN for adaptive predictive control systems could facilitate the development of an automated drug infusion apparatus because it provides robust control even when acute or large perturbations and inter-individual differences in the sensitivity to therapeutic agents occur.