Intelligent and strong robust CVS-LVAD control based on soft-actor-critic algorithm.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Cardiovascular Devices

Background:

• Left ventricular assist devices (LVADs) are crucial for managing ventricular failure.
• Adaptive speed adjustment is essential for LVADs to meet individual patient physiological needs.
• Current control systems may lack the responsiveness required for dynamic physiological changes.

Purpose of the Study:

• To propose a novel physiological control system for LVADs utilizing deep reinforcement learning (DRL).
• To enhance LVAD output regulation by adaptively adjusting device speed based on physiological conditions.
• To compare the performance of the DRL controller against a traditional proportional integral derivative (PID) controller.

Main Methods:

• Developed a DRL-based control system estimating blood flow requirements using a Starling-like method.
• Simulated physiological changes including vascular resistance, myocardial contractility, and rest-to-exercise transitions.
• Conducted single and mixed-factor experiments to evaluate DRL versus PID controller performance.

Main Results:

• The DRL controller achieved 47.6% lower sum of absolute error (SAE) compared to the PID controller.
• The DRL controller demonstrated a 38.6% faster response time than the PID controller.
• LVAD performance showed significantly improved accuracy and speed with the DRL controller.

Conclusions:

• The DRL-based LVAD controller offers superior performance in regulating blood flow.
• This system demonstrates enhanced responsiveness to diverse and dynamic patient physiological needs.
• DRL represents a promising advancement for optimizing LVAD therapy and patient outcomes.