Bladder cancer is a serious disease; early detection and assessment is essential for a good prognosis. When a tumor is detected (e.g. using cystocopy: inspection using a thin steerable instrument with miniature camera), a treatment plan has to be made which may involve removing the tumor or the whole bladder, depending on the tumor stage. To study the muscle invasiveness, optical coherence tomography (OCT) could be used.
The Next-gen in-vivo project aims to develop a robotic cystoscope with miniature camera and OCT sensor to scan the bladder and assess muscle invasiveness of any tumors. In order to properly scan the whole bladder, define the location of any tumors, and especially perform quantitative comparisons of successive sessions, it is useful to make a 3D reconstruction from cystoscopy videos.
Our earlier work allows to make offiline 3D (and 2D) reconstructions from cystoscopy videos using a combination of COLMAP and in-house developed software, utilizing Structure-for-Motion (SfM) / Simultaneous Localization and Mapping (SLAM) algorithms to calculate and iteratively optimize the scene. While it currently produces reasonable results, it is relatively slow, needs several manual steps and can currently only be used offline.
This MSc project aims to implement a workflow which enables real-time 3D reconstruction of cystoscopy videos. Key to success is to make optimal use of readily available libraries to make the assignment doable within the timeframe of the MSc project. The following steps are part of the project:
- Identify scope of project, define research question
- Literature study: current state-of-art of 3D reconstruction of bladder and other hollow organs
- Exploration: try out and evaluate open source frameworks, pipelines, libraries, software packages
- Plan and execute production phase: which software tools, evaluation methods etc
- Evaluation on phantoms and pre-recorded in-vivo cystoscopy videos
- Write report, give presentation