The Endovis Included Four Different Categories: Gastrointestinal Image Analysis GIANA

The Computer Vision and Machine Learning (CVML) Research Group from the School of Engineering participated and came first in two competitions organised as part of the Endoscopic Vision Challenge “EndoVis” (https://endovis.grand-challenge.org/) held at the Medical Image Computing and Computer Assisted Intervention (MICCAI) conference on 10 September 2017 in Quebec, Canada.

The EndoVis included four different categories: Gastrointestinal Image ANAlysis “GIANA”, Surgical Workflow Analysis in the SensorOR, Robotic Instrument Segmentation and Kidney Boundary Detection, with the GIANA challenge consisting of three independent tasks: polyp detection and localization in videocolonoscopy, polyp segmentation in colonoscopy images and angiodysplasia detection and localization in wireless capsule endoscopy images. The university’s CVML group participated in the GIANA’s Polyp Segmentation and the Surgical Workflow Analysis challenges. In both of these challenges the CVML came out on top competing, among others, alongside teams from: Simon Fraser University (SFU); TrueAccord and Massachusetts Institute of Technology (TA-MIT); University College of London (UCL); Japan National Institute of Informatics, Tokyo; Technical University of Munich (TUM) and the German National Center for Tumor Diseases (NCT), Dresden.

For the Surgical Workflow Analysis challenge UCLan teamed up with the University Cote d’Azur (UCA) from France. The team consisted of two UCA MSc students on the ERASMUS+ internship at the CVML lab, Frederic Pecioso (UCA and the UCLan’s visiting professor) and Bogdan Matuszewski (UCLan). The second team, participating in the GIANA’s Polyp Segmentation challenge, was all UCLan team consisting of: Yun Bo Guo (a PhD student at the School of Engineering), Pedro Henriquez and Bogdan Matuszewski.

Winners of the GIANA’s Polyp Segmentation challenge. From left to right: UCLan’s CVML (1st place), SFU ( 2nd place), TA-MIT (3rd place) and UCL (joint 3rd place) (http://www.cvc.uab.es/outreach/?p=618).

The Endoscopic Vision Challenges are concerned with automation of the minimally invasive surgeries and diagnoses. The automation is mainly achieved with a help of cameras used to observe the internal anatomy during diagnostic and surgical procedures. These challenges are instrumental in the development of robust data analysis methodologies advancing the endoscopic image processing and surgical vision, and subsequently their respective translation to the clinical practice.