Markerless registration for image-guided endoscopic retrograde cholangiopancreatography (ERCP)

This paper proposes methods for markerless registration, which enable tracking pose of the endoscope camera in real time for implementation of the image-guided ERCP. Edge-based initialization is developed to determine the initial pose of the endoscope camera. Images of virtual endoscope are rendered from the virtual 3D organ model constructed from the patient's CT images. The similarity between edges on the image of the virtual and real endoscope is exploited for registration. An optical-flow-based tracking method is developed to track the changes starting from the initial pose of the endoscope camera in real time. The redefinition method is proposed to prevent the accumulation of the tracking error. Accuracy of the proposed methods is compared with the previous methods. The initialization method reduces 5.2 mm, 33.1 degrees, and 10.9 degrees of the position, direction, and roll angle error, on average, respectively. The tracking method reduces 3.5 degrees and 1.7 degrees of the hysteresis error in the direction angle and roll angle, respectively, with 15% faster update rate.