A New Control System for Continuous Flow Cavopulmonary Assist Devices Based on an Artificial Neural Network

Continuous flow (CF) cavopulmonary assist devices (CPAD) are currently being developed as the substitute for absent subpulmonary power source in patients with univentricular Fontan circulation. CF-CPAD need to avoid vena caval suction, and autonomously change pump output to meet physiological demand. Ideally, CF-CPAD would also enhance pulmonary vascular pulsatility, which is diminished in patients with Fontan circulation. Pulsatility reduction was related to the endothelial dysfunction of the lungs, elevated pulmonary resistance, and arteriovenous malformations in the lung. A control algorithm was developed for CF-CPAD, which could predict cavopulmonary pressure head (CPPH) with a neural network (NN) model and increase pulmonary vascular pulsatility with a proportional-integral controller that switches between a high and a low CPPH setpoint. Input data of the NN model include CPAD current and speed, the output data are CPPH. The developed NN method was tested in-silico during rest, exercise, and increased vena caval resistance. Various conditions including (1) constant CF-CPAD speed control as baseline, (2) direct measurement of CPPH with pressure sensors, and (3) an extended Kalman Filter (EKF) to estimate CPPH were compared to the performance of this NN method. Results demonstrated that the baseline Fontan supported with CF-CPAD has diminished pulsatility. The proposed method provided physiologic perfusion, avoided vena caval suction, and augmented pulmonary pulsatility by up to ~10 mmHg, compared to ~6 mmHg with EKF method. The proposed NN control algorithm estimated CPPH faster and more accurately than the EKF control strategy, and had a similar performance when CPPH was directly measured.