Catheter and Vessel Shape Estimation for Guidance of Robotic Catheters in Endovascular Surgery

Endovascular procedures are complex minimally-invasive surgical interventions, in which catheters are navigated through the patient's blood vessels. The complexity of these procedures stems from the restricted access to the vasculature: clinicians need not only to resort to external imaging, but also to develop an increased dexterity to precisely manipulate the catheters inserted deep in the patient. Current endovascular procedures predominantly take place under X-ray imaging. In addition to exposing patients and clinicians to ionizing radiations, X-ray imaging only provides a limited 2D visualization where catheters and blood vessels are difficult to localize. Furthermore, conventional catheters are difficult to maneuver because of their intrinsic compliance and their typical manual operation from outside the patient. Advances in robotic technology have fostered the development of robotic catheters that aim to decrease the complexity of endovascular surgery. These robotic catheters embed additional actuators and sensors to better control the catheter tip and/or increase the clinician's and catheter's situational awareness without depending on X-rays. Safe and robust catheter navigation is however still challenging to achieve, as sensor data need to be processed into relevant information that the clinician or robotic catheter can use to adjust the course of their action during navigation. This is especially difficult when operating in the deformable and fragile environment that is the human vasculature. The complexity of endovascular surgery may therefore lead to adverse events caused by acute contact between the catheter tip and the vessel wall. Examples of such events are vessel perforation and embolization of calcified plaques with potential critical neurological consequences. We hypothesize that enhancing the clinician's and robotic catheter's situational awareness significantly helps in preventing contact between the catheter tip and the vessel wall. This awareness is created by i) robustly reconstructing and localizing the catheter shape with respect to the surrounding blood vessel, and ii) simultaneously estimating the vessel geometry. This thesis therefore aims to investigate, develop and evaluate in silico and in vitro probabilistic methods to estimate the catheter and vessel shape from sensory data. A virtual reality (VR) environment is first developed to simulate robotic endovascular procedures. VR provides a configurable environment in which ne