- The workshop proceedings are now available.

Background and Previous Events

Because of their unpredictable appearance and shape, segmenting brain tumors from multi-modal imaging data is one of the most challenging tasks in medical image analysis. Although many different segmentation strategies have been proposed in the literature, it is hard to compare existing methods because the validation datasets that are used differ widely in terms of input data (structural MR contrasts; perfusion or diffusion data; ...), the type of lesion (primary or secondary tumors; solid or infiltratively growing), and the state of the disease (pre- or post-treatment).   

In order to gauge the current state-of-the-art in automated brain tumor segmentation and compare between different methods, we are organizing a Multimodal Brain Tumor Image Segmentation (BRATS) challenge in conjunction with the MICCAI 2014 conference. For this purpose, we are making available a large dataset of brain tumor MR scans in which the relevant tumor structures have been delineated. 

This challenge is in continuation of BRATS 2012 that was held in conjunction with MICCAI 2012 in Nice, and of BRATS 2013 that was part of MICCAI 2013 in Nagoya. For results of the two earlier challenge workshops, please see the recently submitted BRATS manuscript. (Which also serves as the reference for the BRATS data sets available from the Virtual Skeleton Database hosted at Bern University.)

Data, Tasks, and Challenge Format

Task 1:  Lesion Segmentation. Different from the last years we will have a significantly enlarged training and testing data set from the NIH Cancer Imaging Archive (TCIA) with about 300 high- and low- grade glioma cases. Each data set will have T1 MRI, T1 contrast-enhanced MRI, T2 MRI, and T2 FLAIR MRI volumes. Structures have been segmented using manual and algorithmic annotations, with classes as defined in the past years -- essentially comprising the whole tumor, the tumor core (including cystic areas), and the Gd-enhanced tumor core. Annotations are described in the original instructions for the manual annotation of tumor structures, and in the recently submitted BRATS manuscript. Accurately segmenting these structures is the first and primary task of this challenge.

Task 2:  Longitudinal Lesion Segmentation. As a new feature for BRATS 2014 we will indicate those data sets that form a longitudinal sequence of observations for the same patient. More specifically, there will be about 25 data sets with image volumes from 3-10 time points. We encourage participants interested in longitudinal lesion segmentation to make use of this information. To identify the best algorithms for longitudinal brain tumor quantification -- which is of major clinical relevance -- we will offer a second ranking of the participating groups that is only based on the longitudinal data sets. It is possible to only participate in the longitudinal lesion segmentation task.

Task 3:  Diagnostic Image Classification. As we will not disclose the diagnosis of the patient for the test cases (i.e., low- or high-grade glioma), we ask the participants of the segmentation challenge to predict these global diagnostic labels as well (optional). Predictions will be evaluated in a third ranking. It is possible to only participate in the global classification task, or to participate in the global classification task in conjunction with the Brain Tumor Digital Pathology Challenge that is evaluating the same training and test data.

Joint MRI analysis and histology classification tasks. Further information about the NIH Cancer Imaging Archive data, and on options to jointly analyze image features of multimodal MRI and histology scans is summarized on the web page of the Computational Precision Medicine workshop that hosts both challenges during MICCAI 2014.

Participation and Important Dates

Data availability. The co-registered, skull-stripped, and annotated TCIA image data sets will be made available in May-June; the pre-processing will be identical to the processing of the past years. Participants are encouraged to start developing their algorithm using the data from the past challenges hosted at the Virtual Skeleton Database (VSD). The new training data will also be distributed via the VSD. It will be available in early July containing about 400 multimodal data set from about 250 patients. We ill have about 50 test data sets.

Short papers. Participants will have to evaluate their segmentation performance on the training data, and submit a short paper describing preliminary results as well as the segmentation method they are using in early July by the end of July (2-6 LNCS pages; will be reviewed by the organizers - further details to follow). Short papers will be part of the proceedings distributed by the MICCAI organizers.  It is sufficient when they report initial results on the old  BRATS data set.

Evaluation. The independent set of test scans will be made available to each participating team on September 8th. The teams will analyze the images using their local computing infrastructure and will have to submit their segmentation results until September 12th to the VSD submission system. Teams who want to make a full test run of their image processing algorithms can participate in a "leaderboard evaluation" that is held under test conditions during August (participation is optional).

Challenge workshop. Results for all three challenge tasks will be reported in a session that is part of the Computational Clinical Decision Support and Precision Medicine in Brain Cancer pre-conference workshop (S-W17) on Sunday, September 14. Challenge participants will have to register to this full-day event (that also features a session on the related Digital Pathology challenge, and an overview over the NIH Cancer Imaging Archive).

Post-conference journal paper. In the weeks following the challenge the participating teams will be invited to contribute to a joint paper describing and summarizing the challenge outcome, which we will then submit to a high-impact journal in the field.   


Bjoern Menze, TU Munchen / Inria Sophia-Antipolis
Mauricio Reyes, Bern University