Learning Catheter-Aorta Interaction Model Using Joint Probability Densities

Yohannes Kassahun, Bingbin Yu, Emmanuel Vander Poorten

In: 3rd Joint Workshop on New Technologies for Computer/Robot Assisted Surgery. Joint Workshop on New Technologies for Computer/Robot Assisted Surgery (CRAS-13) 3rd September 11-13 Verona Italy Seiten 158-160 9/2013.


Catheter based diagnosis and therapy of cardiovascular diseases is becoming more popular these days. Often the vasculature is being accessed from a less invasive location remote to the cardiac region. For transfemoral approaches of TAVI procedures the vasculature is for example accessed through a cannula inserted into the patient’s groin and then moved gently up to the heart [1]. Due to the complex and deformable nature of both the vasculature and the catheter, the overall controllability of the catheter itself is low in such a case. Tissue damage, dissection or perforation of the vessel and even of the heart cannot be ruled out [2], [3], [4]. The Cognitive AutonomouS CAtheter operating in Dynamic Environments (CASCADE) [5], a recent EU-funded FP7 project, investigates autonomous catheter control and explores machine learning techniques to learn the input-output behavior of the catheter inside vessels of artificial mock-ups. The results from this study should enhance the understanding, the control of catheter motion and interaction patterns also during real interventions. Before learning to control a catheter, different ways of learning the interaction model of the catheter with the aorta should be investigated. These will help to transfer the learned model from the mock-up to the real world in the long run. Since it is not always possible to guarantee safety when applying learning methods, it is important to first investigate different ways of learning the catheter-aorta interaction model and evaluate their failure modes.

Deutsches Forschungszentrum für Künstliche Intelligenz
German Research Center for Artificial Intelligence