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

TL;DR

This paper introduces a new method using deep reinforcement learning to optimize blood flow in membrane oxygenators, improving oxygen transfer efficiency.

Contribution

The study presents a novel active pulsatile flow control method optimized via deep reinforcement learning to enhance oxygenator performance.

Findings

The optimized pulsatile flow profile increased oxygen transfer rate by 20.64%.

The method did not compromise hemocompatibility.

A simplified stacked-plate oxygenator design minimized flow path interference.

Abstract

The membrane oxygenator serves as the core component of extracorporeal life support systems, and its gas exchange efficiency critically influences clinical outcomes. However, gas transfer is predominantly limited by the diffusion barrier within the blood-side boundary layer, where saturated red blood cells accumulate. Current research focuses mainly on static approaches such as optimizing fiber bundle configuration to promote passive blood mixing or modifying material properties, which are fixed after fabrication. In contrast, dynamic blood flow control remains an underexplored avenue for enhancing oxygenator performance. This study proposes an active pulsatile flow control method that disrupts the boundary layer barrier by optimizing periodic flow profiles, thereby directly improving gas exchange. A deep reinforcement learning framework integrating proximal policy optimization and long short-term memory networks was developed to autonomously search for optimal flow waveforms under constant flow conditions. A simplified stacked-plate membrane oxygenator was specially designed as the experimental platform to minimize flow path interference. Experimental results demonstrate that the optimized pulsatile profile increases the oxygen transfer rate by 20.64% without compromising hemocompatibility.

Figures