Learning-based autonomous navigation, benchmark environments and simulation framework for endovascular interventions

Endovascular interventions are a life-saving treatment for many diseases, but they suffer from drawbacks such as radiation exposure and the potential scarcity of proficient physicians. Robotic assistance during these interventions could be a promising support for these problems. Research focusing on autonomous endovascular interventions using artificial intelligence-based methodologies is gaining popularity. However, variability in assessment environments hinders the comparability of different approaches, primarily due to each study employing a unique evaluation framework. In this study, we present autonomous endovascular instrument navigation based on deep reinforcement learning for three distinct digital benchmark interventions: BasicWireNav, ArchVariety, and DualDeviceNav. The benchmarks focus on aortic arch to supra-aortic navigation, representing fundamental large-vessel navigation skills. The benchmark interventions were implemented with our modular simulation framework stEVE (simulated EndoVascular Environment). Autonomous controllers were trained solely in simulation and evaluated in simulation and on physical test benches with camera and fluoroscopy feedback. Autonomous control for BasicWireNav and ArchVariety reached success rates up to 98/100 in simulation and was successfully transferred to the physical test benches with a success rate of up to 97/100. The experiments demonstrate the feasibility of stEVE and its potential to transfer simulation-trained controllers to real-world scenarios. However, they also reveal areas that offer opportunities for future research. Furthermore, this work reduces barriers to entry and increases the comparability of research on learning-based assistance systems for endovascular navigation by providing open-source training scripts, benchmarks, and the stEVE framework.