Surface Mosaics of the Bladder Reconstructed From Endoscopic Video for Automated Surveillance

Abstract

Flexible cystoscopy is frequently performed for recurrent bladder cancer surveillance, making it the most expensive cancer to treat over the patient's lifetime. An automated bladder surveillance system is being developed to robotically scan the bladder surface using an ultrathin and highly flexible endoscope. Such a system would allow cystoscopic procedures to be overseen by technical staff while urologists could review cystoscopic video postoperatively. In this paper, we demonstrate a method for reconstructing the surface of the whole bladder from endoscopic video using structure from motion. Video is acquired from a custom ultrathin and highly flexible endoscope that can retroflex to image the entire internal surface of the bladder. Selected frames are subsequently stitched into a mosaic and mapped to a reconstructed surface, creating a 3-D surface model of the bladder that can be expediently reviewed. Our software was tested on endoscopic video of an excised pig bladder. The resulting reconstruction possessed a projection error of 1.66 pixels on average and covered 99.6% of the bladder surface area.