LV Segmentation Challenge - Cardiac Atlas Project

Information

For some reasons, Google Forms have changed the default settings for all forms to not collect email address. We have fixed this issue. If you submitted a request form without email address and does not get any reply from us, then could you please re-submit your request?

We are sorry for this inconvenience. Thank you.

In 2011, we held a myocardial segmentation challenge at STACOM 2011 workshop. This challenge was aimed to establish a set of ground truth or consensus segmentation derived from participants. In total there were 100 test cases and 100 training cases randomly selected from the DETERMINE cohort. Three semi- and two fully-automated segmentation algorithms participated in this challenge. The consensus contours were generated based on the Expectation-Maximization principle implemented by the STAPLE method.

Contour differences from participants. From left to right: middle, apex, base.

You can read more about this project from [Suinesiaputra et al., 2014].

Data

All challenge data are available for you to download and develop your segmentation method. However, we expect you to return your segmentation results from the test data set so that we can improve our consensus contours. In return, we will perform validation on your segmentation results based on the current consensus contours.

Data description:

Training data size	100
Test data size	100
Contours	Binary mask images (only for training data)
Images	Steady-State Free Precession MRI in short and long axis views
Pathology	myocardial infarction

Participate

Consent

Read carefully and agree on the Terms and Conditions for this challenge, as well as the CAP Data Use Agreement policy document.

Request

Complete and submit the request form below

Download

After approval, you can get a login access to download the data and start developing your segmentation method. Note that you must use the data only for the purpose of this segmentation challenge.

Submit

Segment all cardiac frames short-axis MRI views from the test data set. Then submit your segmentation results to us as binary masks, similar with the training mask images.

Results

You will get results back in terms of performance metrics and you will also be able to download the consensus images from the validation dataset.

Frequently Asked Questions

Can I use the data for other project than the cardiac segmentation?

No. The Data Use Agreement document explicitely specifies that the Data and its derived results can only be used for this Consensus Segmentation project.

Why are the current consensus images not available for download?

The current consensus images are kept to maintain the objectivity of this challenge. You can get the current consensus images only if you have submitted your segmentation results from the test data set.

Can I publish a paper based on the test results?

Yes you can. In your paper you must use the following citations:

1. A. H. Kadish, D. Bello, J. P. Finn, R. O. Bonow, A. Schaechter, H. Subacius, C. Albert, J. P. Daubert, C. G. Fonseca, and J. J. Goldberger. Rationale and Design for the Defibrillators to Reduce Risk by Magnetic Resonance Imaging Evaluation (DETERMINE) Trial. J Cardiovasc Electrophysiol, 20(9):982–7, 2009.

2. A. Suinesiaputra, B. R. Cowan, A. O. Al-Agamy, M. A. Elattar, N. Ayache, A. S. Fahmy, A. M. Khalifa, P. Medrano-Gracia, M.-P. Jolly, A. H. Kadish, D. C. Lee, J. Margeta, S. K. Warfield, and A. A. Young. A collaborative resource to build consensus for automated left ventricular segmentation of cardiac MR images. Medical Image Analysis, 18(1):50 – 62, 2014.

Can I publish a paper using only the training data set and its segmentation images as the reference?

Yes, but with certain conditions. If you have not yet submitted your segmentation results on the test data set, then you are still obligated to do so based on the Terms and Agreement document that you have had agreed when joining this project. Apart from the proper citations, as described above, you must not write the segmentation images that come with the training data set as the ground truth. Rather, you must specify them as semi-automated segmentation images, and refer to the following paper:

B. Li, Y. Liu, C. J. Occleshaw, B. R. Cowan, and A. A. Young. In-line automated tracking for ventricular function with magnetic resonance imaging. JACC Cardiovasc Imaging, 3(8):860–6, 2010.

Request Data

For submission, please contact us by email.