Endoscopy and Computer Vision (EnCoV)

EnCoV means Endoscopy and Computer Vision. This group conducts scientific, clinical and interdisciplinary research in, respectively, computer vision, endoscopy and computer-aided medical diagnosis and intervention.

Computer vision is the part of computer science which studies the automatic interpretation of digital images and videos. EnCoV is especially interested in image registration and 3D reconstruction for monocular image data. These form a set of unresolved challenging problems in terms of the theory and its computational implementation. Two cases may be distinguished: the template-based and the template-free cases. The template is a deformable 3D model of the observed object. The former case is also called Shape-from-Template (SfT). It solves image registration and 3D reconstruction by fitting the deformable 3D model to the input image data. The latter case is called Non-Rigid Structure-from-Motion (NRSfM). By analogy to the rigid case widely studied in the literature, SfT would represent the pose problem while NRSfM would be a direct extension of SfM.

Example of 3D reconstruction of a deformable object, the cap, solved by Shape-from-Template (Bartoli et al, IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015).

Endoscopy is to look inside the body using a camera, called an endoscope. EnCoV is especially interested in two types of endoscopy. The first type is laparoscopy, which is for the abdominopelvic cavity, in gynecology and hepatology. Laparoscopy is part of Minimal Invasive Surgery. The second type is colonoscopy, which is for the large intestine.

Interdisciplinarity arises by developing and using computer vision techniques to facilitate computer-aided endoscopy. This forms a natural coupling because the endoscope is primarily a camera and forms an interface between the doctor and the patient. In this respect, EnCoV researches at the frontier of computer vision and medicine, by developing algorithms and testing them on simulated, phantom, animal and patient models. In laparoscopy, EnCoV's systems aid intervention by combining preoperative data with the intraoperative laparoscope's video stream using Augmented Reality and virtual transparency. In colonoscopy, EnCoV's systems aid diagnosis by providing polyp size measurement and class recognition tools.

Example of EnCoV’s augmented reality laparoscopy system applied to the uterus (Bourdel et al, Surgical Endoscopy, 2017).

Heads