Kidney Boundary Detection Sub-Challenge¶

Part of the Endoscopic Vision Challenge¶

In this sub-challenge we are interested in extracting the border/outer edge of the kidney. This is a useful problem in computer assisted surgery for model based alignment where you try to align the edges of a 3D model of the kidney with the edges in the image. This can then be used to provide the surgeon with information about tumors and vasculature that is hidden beneath the tissue surface. 

We want participants to develop methods which can extract the outer edge of the kidney, while ignoring the other edges in the image. This is a very complex problem as there are many intensity changes in normal surgical scenes which can confuse edge detection methods. To solve this problem we have provided 15x 100-frames video sequences, sampled at 2 Hz, where the kidney boundary is manually extracted by a trained team. For evaluation, each of these 15 sequences has a corresponding 50-frames test set and there are also 5 new 150-frames test sets. Users will be evaulated by how well their extracted contours match the contours picked out by our labelling team. 

The winner of this challenge will be rewarded with a \$1000 prize from Intuitive Surgical.

--

The results and presentation for our challenge can be found here.

A paper on the challenge data 'Kidney edge detection in laparoscopic image data for computer-assisted surgery' can be found here (Hattab et al, 2019). The associated GitHub repo is here.

A follow up work to improve edge definition from boundary detection can found here (Arnold et al, 2021). The associated GitHub repo is here.