Challenge Open Access

# Spontaneous Termination of Atrial Fibrillation - The PhysioNet Computing in Cardiology Challenge 2004

Published: Oct. 8, 2004. Version: 1.0.0

Please include the standard citation for PhysioNet:

Goldberger AL, Amaral LAN, Glass L, Hausdorff JM, Ivanov PCh, Mark RG, Mietus JE, Moody GB, Peng C-K, Stanley HE. PhysioBank, PhysioToolkit, and PhysioNet: Components of a New Research Resource for Complex Physiologic Signals (2003). Circulation. 101(23):e215-e220.

### Introduction

The fifth annual PhysioNet/Computers in Cardiology Challenge focuses on this question:

Is it possible to predict if (or when) an episode of atrial fibrillation will end spontaneously?

Atrial fibrillation (AF) is the most common serious cardiac arrhythmia, affecting more than two million people in the US alone. Unlike venticular fibrillation, which is invariably fatal if it is not interrupted, it is possible for atrial fibrillation to be sustained indefinitely, since the ventricles continue to perform the essential function of driving the circulation, albeit inefficiently. The risks of sustained atrial fibrillation are nevertheless serious, and include strokes and myocardial infarctions caused by the formation of blood clots within stagnant volumes in the atria. Evidence suggests that spontaneously terminating (paroxysmal) atrial fibrillation, or PAF, is a precursor to the development of sustained AF.

Although spontaneously terminating episodes of AF are often very short (perhaps a few seconds in duration), it is interesting to note that longer episodes lasting several minutes also occur. These appear to be very similar to sustained (non-terminating) AF. Subtle changes in rhythm during the final minutes or seconds of such episodes may lead to (or predict) termination of AF. Improved understanding of the mechanisms of spontaneous termination of atrial fibrillation may lead to improvements in treatment of sustained AF. If it were possible to recognize the conditions under which PAF is likely to self-terminate, it might also be possible to intervene in affected individuals to increase the likelihood of self-termination of what would otherwise be sustained AF.

### Organization of the Challenge

The fifth in our annual series of challenges was announced on 23 September 2003 at Computers in Cardiology in Chalkidiki, Greece. At that time, we posted a collection of 80 digitized ECG recordings, the AF Termination Challenge Database, containing labelled training data and unlabelled test data, to support this challenge. To enter the challenge, you will need to:

• develop methods to classify the unlabelled test data
• send your classifications to PhysioNet for scoring no later than Saturday, 1 May 2004 at noon GMT
• send an abstract describing your methods and discussing your findings to Computers in Cardiology no later than 7 May 2004
• (optionally) send up to 4 revised entries per event (see below), no later than Tuesday, 14 September 2004 at noon GMT
• (optionally) submit the source code for your classifier, no later than Wednesday, 15 September 2004 at noon GMT.

Details are below.

If your abstract is accepted, you will be expected to prepare a four-page manuscript (due on Tuesday, 21 September 2004) for publication in the conference proceedings, and you will have the opportunity to discuss your work at the conference. To be eligible for an award, you must submit an abstract and attend the conference.

We invite you to submit the source code for your classifier for possible posting on PhysioNet. One of PhysioNet's major goals is to foster the creation and free dissemination of high-quality software for research on clinically and scientifically interesting subjects. Software contributed in the course of previous challenges has stimulated new collaborations among its authors, and offers rare opportunities to compare the strengths of varied approaches objectively. We will select well-constructed submissions and will post them with full credit to their authors on PhysioNet. We encourage you to participate in this activity as part of the challenge, and we offer additional awards to the authors of the most successful algorithms submitted. A selection of these algorithms will be posted on PhysioNet following the conference.

As in most of our previous challenges, there are two events, and you are welcome to participate in either or both of them. Up to four awards of US$250 will be presented during a plenary session of Computers in Cardiology in September. The top-scoring particpant in each event will receive an award of US$250, and the top-scoring participant among those who have submitted the source code for their classifiers in each event will receive an award of US\$250. Qualified participants may receive more than one award.

Although your initial classifications are due by 1 May 2004, you may attempt to improve your results by submitting a limited number of revised entries, until the final deadline of Wednesday, 15 September 2004. The participants who have achieved the best scores on or before the deadline are the winners of each event.

### The challenge database

The 80 recordings in the AF Termination Challenge Database are each one minute in length (excerpted from longer recordings),and each contains two simultaneously recorded ECG signals. The cardiac rhythm is atrial fibrillation in each case. QRS annotations produced by an automated detector are included for the convenience of those who may wish to study the interbeat interval time series rather than (or in addition to) the ECG signals themselves; note, however, that these annotation sets are unaudited and contain small numbers of errors. Each of the 80 records belongs to one of three groups:

• Group N: non-terminating AF (defined as AF that was not observed to have terminated for the duration of the long-term recording, at least an hour following the segment).
• Group S: AF that terminates one minute after the end of the record.
• Group T: AF that terminates immediately (within one second) after the end of the record.

These groups are distributed across a learning set (consisting of 10 labelled records from each group) and two test sets. Test set A contains 30 records, of which about one-half are from group N, and of which the remainder are from group T. Test set B contains 20 records, 10 from each of groups S and T. The challenge is to identify the group to which each of the test set records belongs.

Group N Group S Group T
Learning set n01, n02, ... n10 s01, s02, ... s10 t01, t02, ... t10
Test set B - 10 records 10 records

### How to enter

1. Develop an algorithm for classifying the recordings in test set A or B (or both). The algorithm must perform this task unaided (manual and semiautomated methods are not eligible).
2. Your entry needs to be prepared in a special text format. For event 1, download this template, and replace the "?" characters with your classifications ('N' or 'T') for each of the 30 records in test set A. For event 2, download this template, and replace the "?" characters with your classifications ('S' or 'T') for each of the 20 records in test set B.
3. Write and submit an abstract describing your work following the instructions on the Computers in Cardiology web site. Please select "Computers in Cardiology Challenge" as the topic of your abstract, so that it can be identified easily by the abstract review committee.
4. (Optional) Send the sources for your classifier by email to PhysioNet. Use the subject line "Challenge 2004 entry source", and be sure to include:
• All sources needed to produce a working version of your classifier (except for readily available standard libraries and header files)
• A note describing how to produce a working version of your classifier (a commented Makefile is ideal), and how to run your classifier
Source files in C, C++, Fortran, or Matlab m-code are preferred; other languages may be acceptable, but please ask first. Do not submit any code that cannot be freely redistributed.

### Important dates

Late submissions will not be accepted.

Saturday, 1 May 2004, noon GMT

Deadline for submission of initial entries to PhysioNet. Please don't wait until the last minute! If you miss this deadline, we encourage you to participate unofficially (your classifications will be scored, but will not be eligible for an award).
Friday, 7 May 2004
Deadline for submission of abstracts for Computers in Cardiology 2004.
Tuesday, 14 September 2004, noon GMT
Deadline for submission of final entries to PhysioNet.
Sunday-Wednesday, 19-22 September 2004
Computers in Cardiology, Chicago, Illinois.

How do I get a password for submitting my entry?

Why did the autoscorer reject my entry?

Valid entries must be in plain text format, as in the templates (see the links above). Don't submit HTML documents, MS Word .doc files, or anything else except plain text; the autoscorer won't like it!

Valid entries must also include a classification for each record in the event that you are entering. There are 30 records in test set A (event 1) and 20 in test set B (event 2). Incomplete entries are rejected.

For each event, you may submit up to five valid entries; any further entries in that event are invalid and will be rejected. Only your top-scoring entry in each event determines your standing.

But I can get five more entries using my friend's email address!

The autoscorer won't recognize that ... but the challenge organizers will. Please respect the spirit of the challenge. As we have advised in previous challenges, if you are tempted to submit many entries in order to discover the correct classifications, try playing Mastermind instead!

How are the scores determined?

The score is the number of correct classifications (so a higher score is always better). The maximum possible scores are 30 for event 1 and 20 for event 2. If there is a tie in any event, the award will go to the first participant to submit a top-scoring entry in that event.

Several records appear to include segments that do not appear to be AF. Is there really AF throughout?

The segments were chosen very carefully and with reference to the entire 24-hour recordings from which they were extracted. In a few cases, there are segments with the appearance of low atrial ectopic rhythm that are in fact AF; this appears to be the case in s09 and t09 (from the learning set). Records a24 (from test set A) and b06 (from test set B) begin with sinus rhythm. Record b09 (in test set B) does not contain sinus rhythm.

Can I enter the challenge using a semi-automated method?

You are welcome to participate unofficially (by submitting results for scoring and by submitting an abstract to Computers in Cardiology), but semi-automated methods are not eligible for awards. Please send a brief note to let us know what you are doing.

Why don't you have a challenge about ...?

Each year, we receive many suggestions for challenge topics. We encourage you to contact us with further suggestions.

### Challenge Results

Over twenty teams participated in this year's Challenge, on the topic of predicting if (or when) an episode of atrial fibrillation will self-terminate.

During September's Computers in Cardiology conference, we presented four awards to eligible participants in this year's challenge. In each event, an overall best award went to the top-scoring team, and a "best open source" award went to the top-scoring team among those who contributed the source code for their entries. The overall award in event 1 was presented to Dieter Hayn and his colleagues, and the overall award in event 2, as well as the "best open source" awards in both events, were won by Federico Cantini and his colleagues.

challenge-2004.jpg shows:

Left to right: Steve Swiryn, George Moody, Simona Petrutiu, Federico Cantini, Dieter Hayn.

We wish to thank all those who participated in the challenge and in the lively and illuminating discussions during the scientific sessions of Computers in Cardiology. Brief descriptions of the methods used can be viewed by following the links in the tables below to abstracts submitted by the entrants for presentation at Computers in Cardiology 2004.

Event 1 (sustained vs. self-terminating AF)

The maximum possible score in event 1 was 30. The top scorers were:

Score Entrant
29 (97%)* S Petrutiu, AV Sahakian, J Ng, S Swiryn
Northwestern University, Evanston, Illinois, USA
28 (93%) D Hayn, K Edegger, D Scherr, P Lercher, B Rotman, W Klein, G Schreier
ARC Seibersdorf Research GmbH
Medical University of Graz, Austria
27 (90%) F Cantini, F Conforti, M Varanini, F Chiarugi, G Vrouchos
CNR Institute of Clinical Physiology, Pisa, Italy
ICS-FORTH, Heraklion, Greece
ICU-CCU Dept,. Venizeleio-Pananeio Hospital, Heraklion, Greece
[Software]
27 (90%) M Lemay, Z Ihara, JM Vesin, L Kappenberger
EPFL - CHUV, Lausanne, Switzerland
27 (90%) F Castells, C Mora, R Ruiz, JJ Rieta, J Millet, C Sanchez, S Morell
Hospital Clinico Universitario de Valencia
Universidaa de Castilla la Mancha, Cuenca, Spain
27 (90%) F Nilsson, M Stridh, A Bollmann, L Sörnmo
Lund University, Sweden
Good Samaritan Hospital and Harbor-UCLA Medical Center, Los Angeles, California, USA

Event 2 (AF terminating in one minute vs. immediately)

The maximum possible score in event 1 was 20. The top scorers were:

Score Entrant
20 (100%)* S Petrutiu, AV Sahakian, J Ng, S Swiryn
Northwestern University, Evanston, Illinois, USA
18 (90%) F Cantini, F Conforti, M Varanini, F Chiarugi, G Vrouchos
CNR Institute of Clinical Physiology, Pisa, Italy
ICS-FORTH, Heraklion, Greece
ICU-CCU Dept,. Venizeleio-Pananeio Hospital, Heraklion, Greece
[Software]
18 (90%) B Logan, J Healey
Hewlett Packard Laboratories, Cambridge, MA, USA
16 (80%) Q Xi, S Shkurovich
St. Jude Medical, Sylmar, CA, USA
16 (80%) D Hayn, K Edegger, D Scherr, P Lercher, B Rotman, W Klein, G Schreier
ARC Seibersdorf Research GmbH
Medical University of Graz, Austria

* The top-scoring entry in each of the two events came from the research group of Steven Swiryn of Northwestern University, who had contributed the data used in the challenge. Although these entries were thus ineligible for an award, Simona Petrutiu and her coauthors did not have access to any information beyond what was available to all of the participants, and we wish to recognize her extraordinary achievement in achieving a near-perfect score in event 1 and a perfect score in event 2.

Four additional teams of participants also described their approaches to the challenge:

AN Esgiar, PK Chakravorty
University of Al Tahadi, Sirte, Libya

P Langley, J Allen, EJ Bowers, MJ Drinnan, EV Garcia, ST King, T Olbrich, AJ Sims, FE Smith, J Wild, D Zheng, A Murray
Medical Physics Dept, Freeman Hospital, Newcastle upon Tyne, UK

LT Mainardi, M Matteucci, R Sassi
Dipartimento di Bioingegneria, Politecnico di Milano, Milano, Italy
Dipartimento di Elettronica e Informazione, Politecnico di Milano, Milano, Italy
Dipartimento di Tecnologie dell Informazione, Universit� di Milano, Crema, Italy

FM Roberts, RJ Povinelli
Marquette University, Milwaukee, WI, USA

### Papers

These papers were presented at Computers in Cardiology 2004. Please cite this publication when referencing any of these papers. Links below are to copies of these papers on the CinC web site.

Analysis of the Surface Electrocardiogram to Predict Termination of Atrial Fibrillation: The 2004 Computers in Cardiology/PhysioNet Challenge
S Petrutiu, AV Sahakian, J Ng, S Swiryn

Prediction of Spontaneous Termination of Atrial Fibrillation Using Time-Frequency Analysis of the Atrial Fibrillatory Wave
C Mora, J Castells, R Ruiz, JJ Rieta, J Millet, C Sánchez, S Morell

Automated Prediction of Spontaneous Termination of Atrial Fibrillation from Electrocardiograms
D Hayn, K Edegger, D Scherr, P Lercher, B Rotman, W Klein, G Schreier

Predicting the End of an Atrial Fibrillation Episode: The PhysioNet Challenge
F Cantini, F Conforti, M Varanini, F Chiarugi, G Vrouchos

Detection of Spontaneous Termination of Atrial Fibrillation
B Logan, J Healey

Predicting Spontaneous Termination of Atrial Fibrillation with Time-Frequency Information
F Nilsson, M Stridh, A Bollmann, L Sörnmo

Electrocardiogram Signal Classification Based on Fractal Features
AN Esgiar, PK Chakravorty

Computers in Cardiology/Physionet Challenge 2004: AF Classification Based on Clinical Features
M Lemay, Z Ihara, JM Vesin, L Kappenberger

A Statistical Feature Based Approach to Predicting Termination of Atrial Fibrillation
FM Roberts, RJ Povinelli

##### Access

Access Policy:
Anyone can access the files, as long as they conform to the terms of the specified license.

##### Corresponding Author
You must be logged in to view the contact information.

## Files

Total uncompressed size: 0 B.

##### Access the files
• Access the files using the Google Cloud Storage Browser here. Login with a Google account is required.
• Access the data using Google Cloud "gsutil":
gsutil -m cp -r gs://challenge-2004-1.0.0.physionet.org DESTINATION
wget -r -N -c -np https://physionet.org/files/challenge-2004/1.0.0/

Name Size Modified
cantini-src
papers