PhysioNet/CinC Challenge Databases

This directory contains specialized databases used as part of the annual series of PhysioNet/Computing in Cardiology Challenges. These databases are categorized as training databases (provided with detailed annotations, for challenge participants to use in testing their submissions, but not used for scoring) and test databases (where the true annotations have been withheld, and challenge participants are scored based on how accurately they can reproduce those annotations.) Note that specialized training databases have not been provided for all challenges, and for many challenges, some or all of the test databases have also been kept hidden from participants.

Challenge 2009: Predicting Acute Hypotensive Episodes
Two test databases are provided, containing a total of 50 multi-parameter recordings from ICU patients, some of whom subsequently experienced acute hypotension.
Challenge 2010: Mind the Gap
One training database and two test databases are provided, containing a total of 300 multi-parameter recordings from ICU patients. In each record, one 30-second segment of a single signal has been withheld.
Challenge 2011: Improving the Quality of ECGs Collected using Mobile Phones
Training and test databases are provided, containing a total of 1,539 twelve-lead ECG recordings. Each training set record is annotated as having “acceptable” or “unacceptable” signal quality.
Challenge 2013: Noninvasive Fetal ECG
Training and test databases are provided, containing a total of 175 abdominal ECG recordings from pregnant women; the training set has been annotated with the locations of fetal QRS complexes.
Challenge 2014: Robust Detection of Heart Beats in Multimodal Data
Two training databases are provided, containing a total of 200 ten-minute segments of multi-parameter recordings collected from a variety of sources, and annotated with the location of each beat.
Challenge 2015: Reducing False Arrhythmia Alarms in the ICU
A training database is provided, containing 750 recordings from ICU patients, each of which caused the bedside monitor to report a severe arrhythmia (asystole, ventricular tachycardia, ventricular fibrillation or flutter, or severe bradycardia or tachycardia.) Each record has been annotated as representing either a true or false alarm.
Challenge 2016: Classification of Normal/Abnormal Heart Sound Recordings
Six training databases are provided, containing a total of 3,240 heart sound recordings collected from a variety of sources. In some cases, a simultaneous ECG signal is also provided. Each subject has been classified as either “normal” or “abnormal” (typically diagnosed with coronary artery disease or heart valve defects), and each record is separately annotated as having “good” or “poor” signal quality.
Challenge 2017: AF Classification from a short single lead ECG recording
A training database is provided, containing 8,528 single-lead ECG recordings (collected with a handheld ECG monitor), and annotated as either normal rhythm, atrial fibrillation, other rhythm, or “too noisy to classify”.
Challenge 2018: You Snooze, You Win
Training and test databases are provided, containing a total of 1,893 multi-parameter recordings (including EEG, EOG, EMG, ECG, and SaO2) collected from sleeping patients. Each training set record has been annotated with the patient's sleep stage over time, and any arousals that they experienced.

Questions and Comments

If you would like help understanding, using, or downloading content, please see our Frequently Asked Questions.

If you have any comments, feedback, or particular questions regarding this page, please send them to the webmaster.

Comments and issues can also be raised on PhysioNet's GitHub page.

Updated Thursday, 3 May 2018 at 15:36 EDT

PhysioNet is supported by the National Institute of General Medical Sciences (NIGMS) and the National Institute of Biomedical Imaging and Bioengineering (NIBIB) under NIH grant number 2R01GM104987-09.