Machine learning guided stimuli-responsive catheter for directional drug delivery and dynamic biliary state recognition

Precise drug delivery in the biliary tract remains challenging due to the dynamic physiological environment and lack of control in existing systems. Here we report a thermo- and pH-responsive semi-permeable catheter with unidirectional drug transport and integrated with machine learning-based environmental state recognition. Addressing the critical challenges of low local drug delivery efficiency and the difficulty of systems adapting to dynamic physiological environments in biliary tract diseases, the catheter adapts its swelling behavior and drug permeability in response to changes in temperature and pH. To achieve precise state recognition, real-time electrical signal data is classified using supervised and unsupervised learning algorithms. We simulated six distinct biliary states and achieved over 95 % accuracy in state recognition using a Random Forest model with Gini-based feature selection. The directional wall design ensured asymmetric diffusion and localized drug release. The research findings demonstrate a system capable of sensing and learning from environmental stimuli, laying the foundation for adaptive biliary tract treatment.