Supervised Monocular Depth Estimation of Laparoscopic Liver Video Sequences using Synthetic Data
- type:Master thesis
- person in charge:
Laparoscopic liver surgery is a minimally invasive surgical technique that offers several advantages over traditional open surgery. It involves smaller incisions, resulting in less pain and scarring for the patient. However, laparoscopic surgery is more technically challenging than open surgery and requires highly qualified surgeons. One option to support surgeons would be an intraoperative navigation tool showing sub-surface structures such as vessels and tumors on the laparoscopic video. For this, registration of preoperative 3D CT model, including vessel and tumor information, with the intraoperative laparoscopic video is required. However, registration of both data is challenging due to the smooth liver surface without many visible structures. One option to overcome the lack of clear features on the liver’s surface would be the use of depth information. Recently, various learning-based methods estimating the depth information have been proposed.
In this work, we focus on supervised monocular depth estimation approaches of laparoscopic video sequences. The aim of this project is to investigate the potential of these methods in the field of laparoscopic liver surgery. For training, synthetic laparoscopic video data including ground truth depth information will be used. Finally, the depth estimation is tested on unseen synthetic data as well as on real laparoscopic video images.