(361m) A Robust, Multi-Model Model Predictive Control Approach to Vagal Nerve Stimulation of the Human Cardiac System. | AIChE

(361m) A Robust, Multi-Model Model Predictive Control Approach to Vagal Nerve Stimulation of the Human Cardiac System.


Adeodu, O. - Presenter, Illinois Institute of Technology
Kothare, M., Lehigh University
Yao, Y., Lehigh University
Mahmoudi, B., Emory University
Vadigepalli, R., Thomas Jefferson University
Gee, M., University of Delaware
Electrical stimulation of the cervical branch of the vagus has been shown in pre-clinical studies to have therapeutic potential in the management of a variety of cardiac disorders including atrial fibrillation (AF). However, the lack of clarity on effective combinations of stimulation parameters and inherent variability across patients have been suggested to explain the mixed outcomes from clinical studies focused on vagus nerve stimulation (VNS). Conceptually, a closed-loop regulation of VNS can resolve these issues. To address this gap, we pursued a computational study aimed at the design of a closed-loop, robust VNS control regulation system that automatically updates the magnitude and frequency of electrical stimulation, based on feedback measurements.

We started with a published model of the closed-loop human cardiovascular control reflex. We updated the model to consider the left atrium as an active compartment with a time-varying elastance to capture its successive roles as reservoir, conduit and pump during a cardiac cycle. Then, we model the onset of AF by (1) the introduction of uncorrelated disturbances that cause atrioventricular dissociation, and (2) a reduction in maximum atrial elastance. To close the VNS control loop, we use data from studies that calibrate VNS using known physiological responses to determine which fiber group (A, B or C) is being stimulated. For a qualitative description of the effect of stimulation frequency, we utilize results of computational studies that model the interaction of physiologically and externally induced action potentials within a mammalian nerve. A non-minimal state space model of past inputs and outputs is then used for controller design.

We demonstrate a robust, multi-model predictive control (MMPC) in a simulation study. We propose this control algorithm as a potentially viable, closed-loop VNS control strategy for the short-term mitigation of the effects of paroxysmal AF. Our study illustrates the design of a computational tool towards the development of personalized neuromodulation therapy.