Results of the Medical Segmentation Decathlon challenge published in Nature Communications

The journal paper describing the Medical Segmentation Decathlon has been published this week in Nature Communications and can be found here. This challenge set out to determine the most generalisable medical segmentation algorithm, including image data from ten different anatomical regions and a variety of modalities. Participants from 19 different teams participated in all phases of the challenge which was hosted on our own Grand-Challenge platform. In the final "mystery" phase, their segmentation algorithms worked on segmenting data from regions never seen during the original algorithm training and development (spleen, colon, hepatic vessels).

This paper is the culmination of a huge effort from the organizers from King's College London, our own Diagnostic Image Analysis Group and multiple other institutes from around the world. The overall challenge winner was the nnU-Net, however the paper includes many other important contributions to the field of segmentation and algorithm generalizability.

We are happy to congratulate the paper's 58 authors on the publication of their work, including Bram van Ginneken, Henkjan Huisman, Geert Litjens and James Meakin from our own group. A fantastic achievement and a must-read paper for anyone involved in medical image segmentation.