Spontaneous Termination of Atrial Fibrillation: A Challenge from PhysioNet and Computers in Cardiology 2004
Massachusetts Institute of Technology
Cambridge, MA, USA
Is it possible to predict if (or when) an episode of atrial fibrillation (AF) will end spontaneously?
Atrial fibrillation is the most common serious cardiac arrhythmia, affecting more than two million people in the US alone. Evidence suggests that spontaneously terminating (paroxysmal) atrial fibrillation, or PAF, is a precursor to the development of sustained AF, which carries significant risks. Improved understanding of the mechanisms of spontaneous termination of AF may lead to improvements in treatment of sustained AF.
This year's PhysioNet/Computers in Cardiology Challenge, the fifth in our annual series of open challenges, invites participants to develop automated methods for discriminating between AF that is about to end and AF that will continue indefinitely. PhysioNet provided free access to a set of 80 one-minute ECG recordings (excerpted from 24-hour ECG recordings) that included non-terminating AF (group N), AF terminating one minute after the end of the recording (group S), and AF terminating immediately following the end of the recording (group T). (AF excerpts chosen as exemplars of "non-terminating AF" were observed to continue for at least one hour following the end of the excerpt in each case). Ten labelled examples of each group were provided as a learning set; the others were divided into two test sets. The test set for event 1 contained 30 records from groups N and T, and that for event 2 contained 20 records from groups S and T. In each event, the challenge was to identify the group to which each of the test set records belongs; a point is awarded for each correct classification. Up to five attempts may be made by each team in each event.
Twenty teams of challenge participants had submitted initial results for scoring by the May deadline; at that time, the most successful algorithms had achieved a score of 29 of a possible 30 in event 1, and a perfect 20 of 20 in event 2. Most participants achieved scores of 21 or better in event 1. These results demonstrate that it is possible to predict if and when AF will terminate with high accuracy. Recognition of the conditions under which AF self-terminates is a first step toward the development of therapeutic interventions that may guide the state of individuals experiencing sustained AF towards self-terminating AF.