An approach to robotic guidance of an uncalibrated endoscope in beating heart surgery

Minimally invasive cardiac surgery is performed under direct vision provided by an endoscope. During cardiac bypass surgery, counterintuitive control and small field-of-view endoscopic visualization is not sufficient to localize all relevant structures of the heart. This work presents a first step towards enhanced endoscopic visualization in cardiac surgery by allowing image based steering of the endoscope without the need of camera calibration. In the envisioned clinical scenario, a surgeon or an endoscopy assistant selects a point or an area in the live endoscope video stream and the robotic system steers the endoscope towards the selected target on a beating heart. Experiments in a simulated setup have shown that the algorithm successfully performs various feature tracking tasks within the requirements. The results from these experiment led to an improvement of the algorithm to enhance the final stage of feature capturing. With the lessons learned in the simulated setup, next series of experiments have been performed in a clinical-like setting on a beating heart phantom. The experiments show that all reasonable target features can be reached within 1.2 s and 7 pixels of accuracy at a safe and conservative robot speed.