Reinforcement learning for navigation in percutaneous coronary arteries interventions

Percutaneous coronary intervention (PCI) is a widely used minimally invasive procedure to treat coronary artery disease. Despite its benefits, PCI relies heavily on the dexterity of the operator and can be highly challenging during navigation. Advances in robot-assisted PCI have shown promise in reducing these risks and enhancing procedural precision. This study presents the development and in silico evaluation of a simulation environment equipped with reinforcement learning (RL) to enable autonomous catheter navigation for PCI, with potential applications in training operators and assisting in real-time procedure guidance. Developed using Unity 3D and Unity ML-Agents, our model utilizes a simulation environment to train RL agents for catheter guidance. The navigation within the coronary arteries was modeled using Unity’s game engine, which allows for realistic catheter movements and collision detection with vessel walls using mesh colliders. The automatic catheterization employs a goal-based binary function, reinforced by a checkpoint reward system that directs the agent's movements toward successful navigation. To evaluate the model, extensive simulation trials were conducted with different movement boundaries to track learning progress and refine training strategies. The results from these in silico trials suggest significant improvements in procedural safety and efficiency, indicated by a reduction in navigation error from an initial average of 0.05 (±0.01) mm to less than 0.01 (±0.002) mm. Cumulative rewards steadily increased, showing a final average of 0.5 (±0.1) mm in reward values. These metrics demonstrate the model’s ability to adapt and optimize its performance for a range of catheter navigation scenarios. However, the evaluation is limited to a virtual environment, and further work is necessary to assess how these findings translate to real-world clinical applications. In particular, integrating this RL-based approach with live PCI procedures will require addressing patient-specific variability, real-time physiological changes, and ensuring safe interaction between the AI agent and human operators during procedures. This study represents an important step toward incorporating AI into cardiac healthcare, but practical implementation in clinical settings will require further investigation, including experimental or clinical validation. Future research should focus on testing the method alongside human operators in controlled clinical environments to evaluate its effectiveness as a real-time guidance tool during PCI.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.