Real-time control of plasma dose delivery in plasma medicine using deep reinforcement learning

This study addresses the clinical need for precise control of plasma dose delivery in plasma medicine, focusing on dose intensity rather than the complex and potentially time-varying parameters of plasma devices (e.g., voltage amplitude, frequency, rise time, gas composition, flow rate, and environmental humidity). A real-time control method for plasma dose intensity based on deep reinforcement learning (DRL) is proposed. This method ensures the stability of the plasma output dose intensity (Equivalent Total Oxidation Potential: ETOP intensity) while effectively adapting to unknown and time-varying environmental changes. The developed plasma dose intensity controller employs a DRL algorithm to interact with the environment, learning optimal control strategies to automatically adjust the ETOP intensity output of plasma devices under varying conditions. In silico simulations demonstrate that the ETOP intensity controller excels in baseline setpoint tracking and effectively regulates plasma dose intensity even when environmental parameters (e.g., humidity) fluctuate. Additionally, the controller simultaneously adjusts multiple parameters to regulate ETOP intensity, with its control range and disturbance rejection ability significantly improving as the number of controlled parameters increases. This research highlights the importance of automatic control of dose delivery in clinical applications of plasma medical devices, toward advancing plasma therapy from laboratory research to clinical practice.