A Deep Reinforcement Learning-Optimized Blood Flow Profile for Enhanced Oxygenation Efficiency in Membrane...

Area of Science:

• Biomedical Engineering
• Fluid Dynamics
• Medical Devices

Background:

• Membrane oxygenators are crucial for extracorporeal life support, but their efficiency is limited by the blood-side boundary layer.
• Current strategies focus on static design modifications, neglecting dynamic flow control.
• Improving gas exchange is vital for better clinical outcomes in patients requiring life support.

Purpose of the Study:

• To investigate an active pulsatile flow control method for enhancing membrane oxygenator performance.
• To develop and implement a deep reinforcement learning framework for optimizing blood flow waveforms.
• To assess the impact of optimized pulsatile flow on oxygen transfer and hemocompatibility.

Main Methods:

• A deep reinforcement learning framework using proximal policy optimization and long short-term memory networks was employed.
• An experimental platform with a simplified stacked-plate membrane oxygenator was utilized.
• Optimal pulsatile flow waveforms were autonomously searched and applied under constant flow conditions.

Main Results:

• The optimized pulsatile flow profile significantly improved oxygen transfer.
• Oxygen transfer rate increased by 20.64% compared to baseline conditions.
• The enhanced gas exchange was achieved without compromising hemocompatibility.

Conclusions:

• Active pulsatile flow control is an effective strategy for enhancing membrane oxygenator performance.
• Deep reinforcement learning can successfully identify optimal dynamic flow profiles.
• This dynamic approach offers a promising avenue for improving extracorporeal life support systems.