Klinikum rechts der Isar Technische Universität München
Abteilung für Neuroradiologie

Large Scale VERtebrae SEgmentation (VERSE) Challenge 2019

Large Scale VERtebrae SEgmentation (VERSE) Challenge 2019


With the advent of deep learning, for the task of vertebral segmentation in computed tomography (CT) scans, a big and varied data is the primary sought-after resource. However, such a public dataset is currently unavailable. With this challenge, we aim to release a large dataset of spine CTs comprised of rich collection of fields of view, spatial resolutions, spinal and vertebral abnormalities, collected over several scanners. Based on this dataset, the task is to assess the performance of spine/vertebrae segmentation algorithms from a clinically-deployable perspective.

Spine segmentation is a crucial step in all applications regarding automated quantification of spinal morphology and pathology. The expected outcome of the challenge is two-fold. (1) To make publicly available a dataset closely representing a clinical scenario in terms of its composition of normal variants, different degrees of degeneration, fractures, and implants. For comparison, this dataset consists of nearly ten-fold data as the segmentation challenge in Computational Spine Imaging workshop (CSI 2014) and xVertSeg challenge in CSI 2016. (2) The semi-automated and fully-automated algorithms submitted for such a diverse dataset need to consider several design decisions that would open new research possibilities in the biomedical image analysis community.


Anjany Sekuboyina (Contact person), Informatics & Klinikum rechts der Isar, Technical University of Munich
Alexander Valentinitsch, Klinikum rechts der Isar, Technical University of Munich
Bjoern Menze, Informatics, Technical University of Munich
Jan Kirschke, Klinikum rechts der Isar, Technical University of Munich
Thomas Link, Radiology, University of California in San Francisco


We will provide 80 training and 40 public test sets of MDCT scans. There will be 40 additional hidden test cases. The training data is already available and the public test set will be released August 1st.


The challenge will take place during MICCAI 2019. Performance will be tested for semiautomatic algorithms on the public test dataset and for automatic algorithms on the hidden test dataset and rated based on Dice Similarity Coefficient (DSC) and Mean Symmetric Surface Distance (MSSD).

Please visit http://verse2019.grand-challenge.org/ for all the details of the challenge.