Make sure to read the rules as well.
I have downloaded the data. How do I read it?
All data is stored in Meta format containing an ASCII readable header and a separate raw image data file. This format is ITK compatible. Full documentation is available here. An application that can read the data is SNAP. If you want to write your own code to read the data, note that in the header file you can find the dimensions of each file. In the raw file the values for each voxel are stored consecutively with index running first over x, then y, then z. The pixel type is short for the image data and unsigned char for the segmentations of the training data.
What do the entries in the result tables mean?
For each test case, five different performance measures are computed. These are the overlap error (OE) in percent, the relative volume difference (VD), in percent, and three symmetric distance measures, all in millimeters. These are the mean absolute difference (AD), the average root mean square surface distance (RMSD) and the maximum distance (MD). Each error measure is translated to a score in the range from 0 (lowest possible score) to 100 (perfect result), by comparing them with typical scores of an independent human observer. Finally, the five scores are averaged to obtain one overall score per test case. These scores are averaged to obtain the score for a system. The details of the error measures and the scoring system are explained here.
How often can I submit results?
In principle, you can upload as often as you want. Note however that all results you submit will appear on the website and every system should be substantially different from previous entries. The differences compared to other systems you have submitted must be evident from the submitted pdf file. In other words, you cannot submit different results using the same pdf file. We are committed to avoid 'training on the test set' and therefore do not want teams to send in a series of results that differ only in the settings of some parameters. For parameter tuning and related experiments, you should use the supplied training data.
Can the results of my system be removed from the website?
Currently, we do not offer the possibility for teams to remove submitted results. If you believe there are good reasons to remove certain results that you have submitted, for example, because you have submitted a new system that makes the old results obsolete, please contact firstname.lastname@example.org.
What must be in the pdf document that is required for every submission?
This document is a paper describing the system that has been used to generate the results in such detail that others can reimplement it, in other words, a standard scientific publication or technical report about your work. Preferably you post your paper to a respectable preprint server such as arXiv. If you have published a paper describing your system, please upload that paper or, in case you are not allowed to have the paper in its original form downloadable from this site, upload a description of it and a reference to the paper. If you have strong reasons why you want to withhold detailed information about your method, please indicate the reasons for this in the pdf file you submit and describe the system only briefly.
Why do I have to provide a pdf document and/or a description of every result I submit?
We believe that it is not too interesting to report here the results of systems whose working is unknown. Therefore we require that a description of each system is provided. It should be a description with enough detail, and if you submit multiple results it must be clear what the differences are compared to your other submissions.
Why can't I download the reference segmentation for the test data and perform the evaluation myself?
From our previous experiences with making data sets publicly available we have learned that if we would release the 'truth' for the test data, groups would perform slightly or vastly different types of evaluations. This may lead to incomparable results between papers that have used the same data. To avoid this, we decided on the current procedure, which makes sure that each system is evaluated in exactly the same way. If you would like to perform a different type of evaluation and the lack of a reference makes it impossible for you to do so, please contact the organizers.
What is the difference between an automatic, semi-automatic and an interactive system?
For each system listed on this site, it is indicated whether it is automatic, semi-automatic, or interactive. When a team submits results, it must indicate to which class the system that generated those results belong.
- An automatic system is fully automatic, that is, it should run without any changes on any input scan, including all test scans.
- If a method requires a seed point to be set, or any parameter that may vary by a user for certain cases, or if different settings have been used for different test cases to obtain good results, or if some pre- or postprocessing was applied that was not exactly the same for all test cases, the system is not automatic. Another way of thinking about this is that if we would ask teams to provide an executable program and we would supply it to the test data, we should get exactly the same results as the ones submitted for automatic systems. Semi-automatic refers to those systems that require some input from a human observer, for some or all cases, but which do not demand extensive editing by a human.
- Interactive systems require extensive editing, and typically have a human observer edit the results until he or she is satisfied with the final outcome. As a result, interactive systems will often yield results that are 'as good as manual'. We realize there is somewhat of a gray zone between semi-automatic and interactive.
Please choose what category you think best fits your system and make sure to describe the degree of interaction needed in the pdf file that describes your system.
Where can I find more information about the data and the competition?
A lot of information is available in the introductory article to the workshop proceedings that can be found here.