A Sensorless Control System for an Implantable Heart Pump Using a Real-Time Deep Convolutional Neural Network.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Cardiovascular Physiology

Background:

• Left ventricular assist devices (LVADs) support heart failure (HF) patients but require sophisticated control systems.
• Current physiological control systems for LVADs necessitate pressure feedback, lacking suitable long-term implantable sensors.
• Automatic adjustment of LVAD speed is crucial for patient stability across diverse clinical scenarios.

Purpose of the Study:

• To develop a novel, sensorless adaptive physiological control system for LVADs.
• To design a real-time deep convolutional neural network (CNN) for accurate estimation of preload using LVAD flow.
• To maintain patient hemodynamics within safe physiological ranges without invasive pressure sensors.

Main Methods:

• A deep convolutional neural network (CNN) was developed for real-time preload estimation based on LVAD flow.
• A sensorless adaptive physiological control system was created using FFDL-MFAC and the proposed preload estimator.
• The CNN model underwent 10-fold cross-validation on 100 patient conditions; the control system was tested on 30 conditions across six scenarios.

Main Results:

• The preload estimator demonstrated high accuracy with a 0.97 correlation coefficient and low RMSE of 0.84 mmHg.
• The sensorless controller effectively mimicked preload-based control, preventing ventricular suction and pulmonary congestion.
• The system successfully adapted LVAD response to changing patient states and physiological demands.

Conclusions:

• A novel CNN-based preload estimator enables accurate, sensorless physiological control for LVADs.
• This approach eliminates the need for additional invasive pressure or flow sensors.
• The developed system offers a promising solution for improved management of heart failure patients with LVADs.