A challenge from PhysioNet and Computers in Cardiology 2008
|Final results are now available: papers presented by participants at CinC 2008, scores, reference rankings and additional details for the Challenge 2008 data set, and the open source software contributed by participants in the open source division of the Challenge.|
The ninth annual PhysioNet/Computers in Cardiology Challenge aims to improve understanding of methods for identification and analysis of T-wave alternans in the ECG.
One hundred years after the first observation of T-wave alternans was reported by HE Hering , the phenomenon is widely understood to be an important indicator of risk of sudden cardiac death. Yet for most of that time TWA was believed to be rare. In 1981, at the eighth annual meeting of Computers in Cardiology, Dan Adam, Solange Akselrod, and Richard Cohen first reported the existence of microvolt-level T-wave alternans , too small in amplitude to be detected visually at standard display scales.
T-wave alternans (TWA) is a pattern in the ECG characterized by two (rarely more) distinct forms of T-waves appearing in alternation, at or above a patient-specific heart rate generally in the range of 90 to 120 beats per minute. Although the mechanisms have not been fully elucidated, a large amount of empirical evidence collected during the past 25 years has demonstrated an association between the amount of TWA, the heart rate at which it appears, and the risk of sudden cardiac death (SCD). In particular, the absence of significant TWA in a patient with congestive heart failure, low ejection fraction, or a recent myocardial infarction is strongly predictive of a low risk of SCD. A positive finding in such a patient, though less specific, may indicate that an implantable cardiac defibrillator would be appropriate, an indication that can be confirmed using invasive testing. Since TWA analysis is performed on the surface ECG, it is an inexpensive and non-invasive test. In clinical applications, TWA analysis can be done as part of an exercise stress test, but there is interest in the research community in using conventional long-term (Holter) ECG recordings to observe TWA in the context of activities of daily living. A review by Armoundas, Tomaselli, and Esperer discusses mechanisms that may account for the associations among TWA and other risk factors for SCD, as well as clinical applications of TWA .
A variety of algorithms for detecting and quantifying TWA have been proposed, employing techniques from linear and nonlinear signal processing such as spectral analysis, complex demodulation, counting zero-crossings in a series of correlation coefficients, periodogram and complex demodulation analysis of T-wave principal components, Capon filtering, Poincaré maps, periodicity transforms, statistical tests, moving averages, maximum likelihood estimators and generalized likelihood ratio tests, and more. For a comprehensive and systematic discussion of methods for TWA detection and analysis, see the review by Martínez and Olmos .
The Challenge Data
PhysioNet has a wide variety of data that might be appropriate for this challenge. Visitors were invited to nominate some of these data for inclusion in the challenge data set, from among those in these databases:
- The Long-Term ST Database, which consists of 86 records, each 21 to 24 hours in duration. The records are digitized 2- or 3-channel long-term (Holter) ECG tape recordings, mostly from subjects who had transient myocardial ischemia. Each record is annotated beat-by-beat, and with respect to ST and rhythm changes, and is accompanied by a detailed clincal summary.
- The PTB Diagnostic ECG Database, consisting of 549 records from 294 subjects, most of which are about 2 minutes long (some are shorter). These are of very high technical quality (12 standard leads and 3 Frank leads, sampled at 1 KHz per signal with 16-bit resolution), mostly post-MI patients but with representation of other types of subjects including healthy controls. Each record is accompanied by a detailed clinical summary. This database was previously used in the 2006 QT challenge, and it might be interesting to relate QT intervals and TWA in the same subjects.
- The MIT-BIH ST Change Database, which includes 23 high-quality records of exercise stress tests, each typically 20 to 40 minutes in length. The recordings each contain two ECG signals sampled at 360 Hz; there are no annotations or clinical correlates available.
- The Sudden Cardiac Death Holter Database [alternate link], a collection that currently includes 23 complete 2-channel Holter recordings (7 to 25 hours each) of patients who experienced sustained ventricular tachyarrhythmias and cardiac arrest during the recordings. The records have been digitized at 250 Hz per signal. Twelve records have been meticulously annotated beat-by-beat; algorithm-generated beat annotations are available for the others. No clinical correlates are available.
- The BIDMC Congestive Heart Failure Database, 15 2-channel Holter recordings each about 20 hours in duration, from patients with severe (NYHA class 3-4) congestive heart failure. The records have been digitized at 250 Hz per signal. Age, gender, and NYHA class are available for each patient.
A good challenge data set should include not only clear-cut cases, but also a sampling of cases that might be expected to pose problems (for example, because of changes in conduction patterns, ectopy, or noise).
Participants were asked send a list of the records they wished to nominate for inclusion in the official challenge data set by email to email@example.com with the subject "Challenge 2008 data nomination", no later than noon GMT on Thursday, 17 April 2008. Those who wished to nominate records other than those in the databases listed above were asked to state briefly why they believed their choices are particularly appropriate for this challenge. Participants also were invited to contribute data to PhysioNet to be considered for this challenge, and several did so. Thanks to all who participated in this process.
We have compiled all nominations and selected the most promising cases for inclusion in the official data set, together with new cases that have not been available previously. The official PhysioNet/Computers in Cardiology Challenge 2008 data set is now available (as individual files) here; for convenience, it is also available as a tarball (see instructions here for unpacking tarballs).
If you downloaded the data set before Monday, 21 April 2008, you can bring your copy up-to-date using the update tarball.
Reporting Results and Scoring
- Download an entry form.
- For each record in the challenge data set, participants must detect or estimate the peak magnitude of T-wave alternans using a fully-automated method, and enter that estimate on the entry form. If your method detects but does not quantify TWA, enter 1 as the estimate for each record in which TWA was detected, and 0 as the estimate for the remaining records. Valid entries must include a TWA estimate for each record in the challenge data set.
- Use your favorite text editor to fill in the entry form. Please note that the entry form is a plain text file, not a word processor document; keep it in plain text format. Be sure that your entry form includes the email address where you wish to have your scores sent.
- Send the filled-in entry by email to firstname.lastname@example.org, with the subject line "Challenge 2008 entry". Entries will be accepted until noon GMT on 1 September 2008.
Scores will be determined by the following procedure, once a minimum number of entries have been received:
- For each entry, the records will be ranked by the magnitudes of the associated TWA estimates. Thus the record with the lowest TWA estimate in a given entry receives the rank of 1 for that entry, the record with the second-lowest TWA estimate gets a rank of 2, etc.
- Each record receives a median rank, which is the median of the ranks assigned it by the entries, and a reference ranking is made by sorting the median ranks (i.e., the record with the lowest median rank gets a reference rank of 1, etc.)
- The score for each entry is the Kendall (tau) rank correlation coefficient between the entry ranking and the reference ranking, where 1 is perfect agreement and -1 is perfect disagreement .
You will receive your score by return email. The first scores will be sent on or about 27 April 2008; scores for entries received after that date will usually be sent within a day of receipt.
The reference rankings are initially determined by the entries
received. Final reference rankings will be compiled in August.
Scores reported before that time are preliminary and are subject to
change. Final scores used to determine the winners of the challenge
will be sent to all participants shortly after the deadline for entries
(1 September 2008).
Entering the open source division
As in previous years, the Challenge will include an open source division. You may enter the open source division by sending your source code by email, before noon GMT on Monday, 1 September 2008, to PhysioNet. Use the subject line "Challenge 2008 entry source", and be sure to include:
- Your name and email address
- All sources needed to produce a working version of your software (except for readily available standard libraries and header files)
- A note describing how to produce a working version of your software (a commented Makefile is ideal), and how to run your software
Each source file submitted should begin with a comment block containing the names of its authors and a reference to the open source license you have chosen for it, if any; for example:
/* twarng.c - Detect and measure T-wave alternans using a random number generator Copyright (C) 2008 Herman Foobar <email@example.com> This software is released under the terms of the GNU General Public License (http://www.gnu.org/copyleft/gpl.html). */
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. Following the conclusion of the Challenge, selected entries will be posted, with full credit to their authors, on PhysioNet.
Events and Awards
If you submit an abstract describing your work on the challenge no later than 1 May 2008 to Computers in Cardiology, and a challenge entry no later than 1 September 2008 at noon GMT, you may be eligible for one or more awards that will be presented during the final plenary session of the conference on Wednesday, 17 September 2008.
The Board of Directors of Computers in Cardiology provides an annual grant of US$1000 for challenge awards. This year, a matching contribution from Electrogram, Inc. has allowed us to double the prize fund. We are deeply grateful for the generosity and enthusiastic support of Electrogram and Computers in Cardiology.
An award of US$1000 will be received by the eligible participant who achieves the best final score overall in this year's Challenge.
An additional award of US$1000 will be given to the winner of the open source division. Participants in the open source division are eligible for both awards.
Frequently Asked Questions
If your question is not answered below, please consult the PhysioNet FAQ.
Regarding the estimate of the peak T-wave alternans magnitude, in what units it should be expressed? Does the estimate include the T-wave magnitude or only the added alternans?
You may use whatever units you wish. Estimate the magnitude of the alternans, not the magnitude of the T-waves. (Some methods involve expressing alternans magnitude or power as a fraction of total T-wave magnitude or power; it's fine to do this, or not to do this, as you wish -- but don't add a T-wave measurement to your alternans estimate.)
The scoring is based on how your estimates rank the records (from least to most TWA) compared with a reference ranking of the records. Thus the units you choose don't matter, so long as you use the same units for all of your measurements.
This is a hard problem, and I won't be able to get final results in time for the abstract deadline. What needs to be submitted in the abstract?
Briefly, you will need to submit a short (~300 words) abstract, which may not include illustrations, by 1 May. Abstracts must be submitted on-line.
It would be ideal if you have submitted rough results for the Challenge problem no later than 27 April, so they can be scored and so that you can report your preliminary score in your abstract. If this is not possible, please discuss in your abstract relevant studies you have performed to date, touching on any novel aspects of your methods and quoting whatever results you have. Generally abstracts containing only "we will show ..." without any results are not accepted, so it is very important that you demonstrate in the abstract that you have actually accomplished something already. Given the complexity of the challenge and the short time available for working on it, the abstract reviewers will certainly make some allowance for those who are unable to obtain results on the challenge problem by the abstract deadline, if it appears they will be able to do so before the conference in September.
May I revise my entry?
You may submit entries at any time until the final deadline of noon GMT on 1 September. Each entry will be scored, and you may attempt to improve your score by submitting up to four revised entries (five in all).
You may choose any of your (up to five) entries as the basis for ranking, by sending an email specifying your choice to firstname.lastname@example.org, on or before noon GMT on 1 September. Please remember that it may take up to 24 hours after submitting an entry to receive scores, so try to submit your last entry at least a day before the deadline if you think you may want to exercise your choice. Your last entry will be used for ranking unless you specify otherwise.
If my abstract is accepted, what else must I do?
In September, if your abstract is accepted, you will be expected to submit a four-page paper, which may be illustrated, for publication in Computers in Cardiology on-line and in print. You will also be expected to attend the conference (14-17 September 2008 in Bologna, Italy) and to present your work in one of the scientific sessions of the conference, either as a poster or as a 10-minute oral presentation. Your paper and presentation should include your final results.
Why don't you have a challenge about ...?
Each year, we receive many suggestions for challenge topics. We encourage you to write to us with further suggestions.
- Hering HE. Das Wesen des Herzalternans. Münchener med Wochenschr 1908;4:1417-21.
- Adam DR, Akselrod S, Cohen RJ. Estimation of ventricular vulnerability to fibrillation through T-wave time series analysis. Computers in Cardiology 1981;8:307-310.
- Armoundas AA, Tomaselli GF, Esperer HD. Pathophysiological basis and clinical application of T-wave alternans. JACC 2002;40:207-217.
- Martínez JP, Olmos S. Methodological Principles of T-wave alternans: a unified framework. IEEE Trans Biomed Eng 2005;52(4):599-613. [requires subscription]
- Abdi H. "The Kendall Rank Correlation Coefficient", in Salkind N (Ed.) Encyclopedia of Measurement and Statistics. Thousand Oaks (CA): Sage (2007).