A Dynamic Model of Multi-state LVAD Based on LSTM Neural Network

Left ventricular assist devices (LVADs) assist cardiac function by regulating pump speed to support blood circulation. Traditional models of the heart and LVADs often rely on static equivalent circuit representations. However, the dynamic nature of physiological parameters across different human body states can lead to suboptimal outcomes with constant speed control strategies. This paper presents a novel approach to address this challenge by proposing dynamic circuit models for both the heart and LVADs. Leveraging Long Short-Term Memory (LSTM) neural networks trained on historical data of human blood pressure and blood flow, our method captures intricate patterns in individual physiological states. Simulation results demonstrate the efficacy of the proposed algorithm in accurately representing various states, such as non-suction and suction. The results show that the proposed dynamic model reduced the error of the LVAD model.