Software tools to enable 3D bladder reconstruction from clinical white light cystoscopy videos

Abstract

Bladder cancer is the tenth most common cancer in the world and has a 31-78% five-year recurrence rate. This high recurrence rate drives the need for individuals diagnosed with bladder cancer to receive cystoscopies every three months to one year for the rest of their lives. White light cystoscopies are the gold standard for diagnosis and follow-up in the bladder; during a cystoscopy, the clinician views the inside of the bladder with a camera passed through the urethra and into the bladder. Clinicians often use cystoscopy data to inform surgery and treatment; however, cystoscopy videos are cumbersome to review, so clinicians currently save a few notes and frames then discard the rest of the video. This video data can instead be saved in the form of a 3D bladder reconstruction, which preserves the texture and shape of the bladder in a form that is informative and convenient-to-review. Reconstructions, however, are inherently limited by the quality of their underlying data. This dissertation aims to improve reconstruction quality by improving the quality of the underlying video. We first improved reconstruction quality by removing a honeycomb-like artifact associated with equipment often used in clinical settings. Next, we developed a real-time pipeline to detect frames likely to inhibit reconstruction and give the clinician the opportunity to recollect the past few seconds of data. Finally, we developed a cystoscopy simulator to serve as a platform for testing and development of new clinical tools. In summary, this dissertation helps to improve reconstructions from clinical cystoscopies and aids development of future tools, which helps to enable the use of 3D bladder reconstructions in clinical settings to improve patient care.