Learning Based Sim-to-Real Autonomous 3D Navigation for Robot-Operated Right Heart Catheterization

Cardiovascular Disease (CVDs) were responsible for 20.5 million deaths in 2021, comprising approximately one-third of all global fatalities and maintaining their position as the leading cause of human mortality. 1 Right Heart Catheterization (RHC) stands out as a highly effective clinical diagnosis for some conditions of CVDs. The procedure involves the use of a specialized Swan- Ganz catheter, which is inserted through a minor incision in either the patient’s femoral or jugular vein. It is then meticulously navigated into the inferior vena cava (IVC), right atrium (RA) and right ventricle (RV), and precisely positioned within the pulmonary artery (PA). Numerous research have been launched to advance the de- velopment of robotically operated cardiac interventions, with the goal of reducing the physical strain on clinicians associated with manual procedures. Additionally, there is a growing interest in the integration of machine learning (ML) and artificial intelligence (AI) algorithms for enabling autonomous functionality in interventional robots. 2, 3, 4 For example, Chi et al. achieved tra- jectory optimization for catheter interventions in various cardiac phantoms through Learning from Demonstrations (LfD) using Gaussian mixture models (GMM). 5 A new trend is emerging, focusing on the use of virtual environments for policy learning and training, with seam- less transfer from simulation to real robotic autonomous operations (Sim-to-Real). Y. Cho et al. successfully implemented Percutaneous Coronary Intervention (PCI) using a Behavioral Cloning algorithm, where policies were initially trained in a simulation environment before being applied in real-world 2D autonomy scenarios. 6 This paper presents a novel, learning-based robotic system designed for 3D navigation of a catheter, specifically tar- geting Right Heart Catheterization (RHC). Experiments demonstrate that the transfer from simulation to the real- world can be achieved by using Behavioural Cloning (BC) algorithms, which in turn enable autonomous robotic operated interventions within patient-specific phantoms.