Real Time Localization of Cystoscope Angulation in 2D Bladder Phantom for Telecystoscopy

Abstract

Telecystoscopy can lower the barrier to access of critical urologic diagnostics for patients around the world. A challenge to robotic control of flexible cystoscopes and intuitive teleoperation is estimation of the pose of the scope tip. We demonstrate real-time localization using video recordings from prior cystoscopies and 3D reconstructions of the patient’s bladder to estimate cystoscope angulation We map prior video data into a low dimensional space as a dictionary so that new images can be matched to a nearest neighbor. The cystoscope position is estimated by the current image’s relationship to a matched dictionary frame. Frequent cystoscopies are necessary in post-treatment surveillance of bladder cancer patients, and video from previous procedures may be available during telecystoscopy. A cystoscope with servo-controlled angulation was inserted into a 2D + height bladder shape with a panorama of a urothelium on to the inside. Scans of the surface were performed with: varying speeds; different fields of view; and bladder tumors inserted into the panorama physically and digitally. Videos were used to create 3D reconstructions, dictionary sets, and test data sets for analyzing our algorithm’s computational efficiency and accuracy compared with a SIFT-only localization. Our algorithm found a nearest neighbor image in 96-100% of frames in under 60ms per image compared to SIFT’s ability to find an image match in 56-84% of frames in more than 6000ms per image. Our algorithm, with a first stage rate of nearly 20 Hz, is a promising tool for real-time estimation of tip location in robotic cystoscopy when prior cystoscopy images are available.