Multi-objective Neural Predictive Control for Biventricular Assist Device: Balancing Circulatory Volume with Preload-Based Control

Abstract

Rotary blood pumps are essential for providing mechanical circulatory support to patients with heart failure. However, maintaining pulmonary and systemic circulatory volume balance when using two rotary blood pumps for biventricular support is challenging. In this paper, we propose a novel approach to address this issue by combining a multiobjective neural predictive controller (MONPC) with a preload-based Frank-Starling-like controller (PFS) for a dual rotary blood pump biventricular assist device. We evaluate two different configurations: PFSL-MONPCR and MONPCL-PFSR. The PFS controls the flow rate of one pump based on preload, while the MONPC adjusts the other pump to meet cardiac demand, prevent pulmonary congestion, and avoid ventricular suction. We compare the performance of these controllers with a Dual Independent Frank-Starling-like control system (DI-FS) and a constant speed controller through numerical simulations. The results demonstrate that MONPCL-PFSR effectively unloads the congested left ventricle while maintaining high cardiac output during exercise. In contrast, improper flow regulation by DI-FS leads to pulmonary congestion. Moreover, during blood loss, PFSL-MONPCR exhibits the lowest suction risk compared to the constant speed mode, which results in negative right ventricular preload. Furthermore, when considering sensor noise and time delays in the signals, the proposed controllers demonstrate robustness during the transition from rest to exercise. Our study highlights that the proposed controllers effectively match pump flow with cardiac demand, ensuring hemodynamic stability in biventricular assist devices.