Target vessel identification in endoscopic coronary artery bypass grafting

Minimal invasive direct coronary artery bypass grafting (MIDCAB) procedures are potentially advantageous over conventional open surgery, while one major limitation is the narrow view of the endoscope and the difficulty in detecting stenosis from the endoscopic view. In this paper, we propose a frame to align the endoscope view and the 3D-vessel model reconstructed from angiography. We register the endoscopic video frame with the 3D-vessel model by electro-magnetic sensor, and then we segment the vessel in the endoscopic video frame with HSV (hue, saturation, value) and hessian transformation, and refined the registration with the vessel model by Coherent Points Drifting. We have validated the proposed method on a static phantom heart model and a real patient's endoscope video. RMSE (root mean-square error) was employed to evaluate our method and the registration error for the phantom model were 1.1+-0.2mm, for the real patient was 3.1mm.