Optimising dry powder inhalers for lung deposition with a DRL CFD coupled framework

Respiratory tract infections remain a significant global health challenge, with inhalation therapies playing a critical role in targeted drug delivery. This study investigates the application of Deep Reinforcement Learning (DRL) to determine the physical parameters of inhalers, aerosol particles, and their dissemination flow that maximise lung deposition for oral inhaled drug delivery. Results from this study suggest that DRL implemented with Proximal Policy Optimisation (PPO) works well for highly multivariate cases. In this study, the DRL performance in cases designed with the maximum number of variables (seven parameters) produced deposition efficiencies (DE_lung) that were 7.7% higher than those achieved using response surface methodology (RSM). The results also indicated the combination of physical parameters plausible to obtain a DE_lung of 95.8% in the United States Pharmacopeia (USP) throat and 89.6% in the Virginia Commonwealth University (VCU) throat. The results showing that DRL outperformed RSM in all optimisation tasks with the cost of running extra simulations. These findings highlight the potential of DRL for optimising inhalation therapies and contrast this approach with RSM, demonstrating their respective contributions to enhancing aerosol delivery systems.