PhysioNet/CinC Challenge 2002 Event 2: Classification


This page contains information about how to participate in event 2. For background on this challenge, see the challenge overview. The complete set of rules for event 2 can be found here.

As in previous challenges, there are two events, and the top-scoring participant in each event will receive an award. Event 1 required participants to write a program that can generate a realistic 24-hour time series of inter-beat (RR) intervals. The outputs of these programs form a part of the challenge dataset used for event 2, in which participants attempt to sort the data into real and synthetic classes.

About the challenge dataset

The PhysioNet/CinC Challenge 2002 dataset consists of 50 time series of interbeat (RR) intervals, designated rr01, rr02, ..., rr50. Each series contains between 20 and 24 hours of data (between 70,000 and 130,000 intervals). Approximately half of the series (those belonging to group A) contain data derived from ambulatory ECG recordings of subjects between the ages of 20 and 50 who have no known cardiac abnormality (although, as for most healthy adults, they may have rare isolated ectopic beats, which have not been filtered out of these time series). The other half (those belonging to group B) contain synthetic data, which have been generated using the models submitted by participants in event 1 of the challenge, and by one or more models provided by the challenge organizers. Each model has been used to generate two time series.

The object of event 2 is to determine which of the series belong to group A (the real data), and which to group B (the synthetic data).

Special thanks to Diane Perry, Isaac Henry, and Joe Mietus for their assistance in assembling the group A data.

Downloading the challenge dataset

The challenge dataset is available in two formats:

Preparing your entry

After downloading the challenge dataset, study it. If you are not familiar with the characteristics of real interbeat interval time series, it may be helpful to look at the MIT-BIH Normal Sinus Rhythm Database, which is a source of real data with all of the characteristics of the time series in group A. The interbeat intervals from the 18 records of this database, which may be useful as reference material, are contained in nsrdb-rr.tar.gz (1633730 bytes, in the same format as dataset.tar.gz).

Classify each series in the challenge dataset as A (real), B (synthetic), or C (unknown). You will receive 2 points for each correct classification, and your score will be reduced by 1 point for each incorrect classification. Your score will be neither increased nor reduced for any series that you classify as C (unknown). Possible scores thus range from -50 to 100.

Your entry needs to be prepared in a special text format. Download this template, and replace the `?' characters with your classifications (`A', `B', or `C') for each of the 50 series in the challenge dataset.

Submitting your entry for scoring

Please double-check your entry before submitting it to be sure it is in plain text format, exactly like the template file. HTML-formatted entries will confuse the automated scorer and will not be scored.

To submit your entry successfully, you will need:

After filling in your e-mail address and password on the entry form. enter the name of the file that contains your correctly-formatted entry (you may use the Browse button to find it), and click on Upload entry.


Please be careful not to submit more than one copy of the same classifications, since each submission will be counted against your quota of submissions. If this happens inadvertently, please send a note quoting the date and time of the duplicate submission; your submission counter will be adjusted manually if the error can be confirmed.

If you are tempted to try submitting many entries in order to learn about the correct classifications, why not play Mastermind instead? We will reject obvious attempts to circumvent the spirit of the challenge in this way. Delivery of your score may be delayed if you submit more than one entry per day.

UNIX versions of Netscape, and possibly other browsers, will allow the user to specify a directory name in the File to upload field of the entry form, but the browser will not transmit anything in this case and will hang until the user interrupts it by clicking on Stop. This behavior is a bug in the browser and cannot be detected or avoided by the PhysioNet web server. Other browsers will detect and disallow attempts to upload directories.