(246a) Model Predictive Control of Vagus Nerve Stimulation in the Rat Cardiac System Using Long Short-Term Memory Network

Cardiovascular disease remains one of the leading causes of mortality worldwide despite several pharmaceutical treatments, motivating a search for a more efficacious therapy. In recent years, vagus nerve stimulation (VNS) has emerged as a potential therapy to alleviate the pathological conditions associated with the cardiovascular diseases. Efforts to successfully apply VNS in clinical trials have been met with challenges surrounding the selection of optimal VNS parameters to characterize the efficacy in treating cardiovascular diseases. Optimization of the VNS electric pulse from the physiological measurements in a closed-loop provides a promising solution to this challenge. In this work, we have developed a closed-loop optimal control approach that could be used to optimize the VNS parameters for a desired physiological response in the cardiac system. Particularly, we developed a long short-term memory (LSTM) neural network model to map the VNS parameters such as the stimulation pulse width and frequency on the cardiac physiology such as the heart rate and mean arterial pressure. We trained our LSTM model on the input-output timeseries synthetic data generated using a published biophysiological mechanistic model of the rat cardiac system. We then deployed this trained LSTM in a closed-loop model predictive control (MPC) framework to optimize the VNS parameters. We tested our closed-loop LSTM-MPC framework in controlling the heart rate and mean arterial pressure predicted from the physiological model of the rat cardiac system and systematically investigated the effect of model mismatch, choice of cost function, and controller design parameters on the overall closed-loop performance. We found that our controller design was able to drive the system to the desired set points under various conditions. Overall, our work demonstrates a proof-of-concept of using LTSM in designing closed-loop MPC strategy to optimize VNS parameters for controlling cardiac physiology.