Currently, visual inspection of the bladder wall for locating bladder cancer is done during a cystoscopy using manually operated instruments. Flexible endoscopes do offer a full range of vision of the bladder but lack dexterity. The conventional tumour documentation is based on provisional hand-made drawings to mark locations and is susceptible to inaccuracy due to variations in drawings and interpretation.
To better direct the treatment and reduce these harmful side effects, the documentation method needs to be improved. In a joint research effort between RadboudUMC and the University of Twente (UT), the use of a CTR and optical coherence tomography (OCT) sensor to inspect the bladder wall is proposed. Improving tracking of the bladder wall texture and classification of suspected tumours across successive cystoscopy sessions.
This thesis will report on the feasibility of using a 2D video to act as a state estimator that can be utilized for applying feedback on the CTR control loop. To detect motion, the video frames, which are individual images, need to contain enough point features. Therefore, the presence and quality of different feature types in bladder footage will be evaluated.
The feature types evaluated will be: SURF (blobs), SIFT (blobs), ORB (corners), KAZE (blobs), BRISK (corners). These feature types are selected from the types supported by MATLAB, the program used for image analysis.