In this thesis, a new robotic platform for endovascular interventions - CathBot (developed at the Imperial College London) - is presented and a vision-based guidance system is developed to offer visual and haptic feedback to the operator, providing information on the interactions between vasculature and the whole catheter during surgery.
The imaging system proposed works on X-ray angiograms and with the help of state of the art machine learning algorithms it is able to detect both the blood vessels and the surgical instruments. The proposed network architecture was able to detect the vasculature with an average accuracy of 94%, and the instrumentation with an average accuracy of 88%. The system achieved 4 frames per second on a mid-end machine. A framework for contact point detection and force estimation is also proposed.
To join the presentation via Microsoft Teams click here