PhD : "Computer Vision Aided Diagnosis and Guidance in Endoscopic Urology"

Ph.D defense: Jorge Lazo

Abstract :
Endoscopy is a minimally invasive procedure used for the detection, diagnosis, and treatment of diseases in hollow organs such as the bladder, the colon the esophagus, etc. In the specific case of the urinary system, it consists of the passage of a ureteroscope through the urethra and bladder, and in case of being necessary, up to the ureter and the kidneys. The visual information obtained from the endoscopic camera helps clinicians with two main tasks: navigation and diagnosis. In this regard, the goal of this Ph.D. project is the development of computer vision systems suitable to be used in endoscopic urology with a focus on the two main purposes for which endoscopic information is used during this procedure: visual information useful for navigation and tissue information necessary for diagnosis.  In the first part, we focus on the task of lumen segmentation, we propose the use of spatial-temporal ensembles to achieve a better generalization of ureter lumen segmentation that is robust enough to perform in tasks with poor visibility conditions, caused by sporadic bleeding, or specular reflections. In the second part, we focus on the task of bladder tissue classification. We propose a new method for bladder tissue classification with a focus on bladder cancer identification, in scenarios where labeled data is limited to only one domain of the two which are usually used in the procedure (NBI and WLI), and there are no identical equivalent pairs for every image on each domain. Finally, we show that the proposed models can be integrated into flexible robots in order to automatize certain tasks. A synergic solution for intraluminal navigation is proposed which consists of a 3D printed endoscopic soft robot and a visual servoing control method based on a lighter version of the segmentation previously proposed. The proof of concept is capable of performing autonomous intraluminal navigation in narrow luminal structures. The proposed robot is validated in anatomical phantoms in different path configurations and conditions different than the ones the model was originally trained on.

Date :   February 20th at 15:30
Place :  BIO1 Room Building 21 DEIB Department Politecnico di Milano Campus Leonardo, Milan, Italy

The meeting can also be joined online at the following link in TEAMS: