Test dataset for: "Automated diagnosis of atrial fibrillation in 24-hour Holter recording based on deep learning:a study with randomized and real-world data validation"

Published: 4 October 2021| Version 4 | DOI: 10.17632/44htzjcgsz.4
, Peng Zhang


This test dataset comprised of 800 24-hour Holter recordings from the test set of the randomized clinical cohort in the paper. Each recording includes the RR-interval data that are extracted from a 24-hour dynamic 12-lead ECG recording captured by a Holter machine (DMS Holter Company, Stateline, NV, USA) at three campuses (Main Campus, Optical Valley Campus, and Sino-French New City Campus) of the Cardiac Function Examination Center (Division of Cardiology, Tongji Hospital, Huazhong University of Science and Technology, Wuhan, China). All the data were initially interpreted by primary cardiologists, then were further reviewed by three senior board-certified cardiologists to ensure the correctness of the base diagnostic labels. The cardiologist committees discussed by consensus the annotated records and provided a reference standard for model evaluation. Each atrial fibrillation(AF) episode included the accurately labeled start time and end time for patients with paroxysmal AF (PAF). The start and end times of each PAF episode were the corresponding time of the first atrial wave with an atrial rate greater than 350 beats/min and the corresponding time of the first P-wave with sinus rhythm after the termination of AF. Specifically, PAF episodes lasting more than 30 s were accurately labeled, and those lasting less than 30 s were labeled as accurately as possible. Moreover, each interval of premature beat or tachycardia was marked as “A” (atrium event) or “V” (ventricle event), and the long RR interval caused by QRS wave dropping was marked as “B” to further label Second-degree atrioventricular block. The 800 recordings included 200 whole-course AF (WAF), 200 PAF and 400 NAF recordings. For WAF patient data, whole recording included only AF signals. For NAF patients, the entire recording data had no AF signals but included normal sinus rhythm, sinus arrhythmia, atrial arrhythmia, ventricular arrhythmia, atrioventricular block and so on. The data of PAF patients included both AF episodes and NAF signals. Both WAF and PAF patients were considered to be AF patients. The type of each recording is indicated by its file name and the comments on the contents can be seen in the picture "Note.jpg". We have uploaded all the 800 files to the system, however, for unknown reasons, only 300 files can be downloaded. We have contacted with the technical support of the Mendeley to solve this problem, if you want to use the full 800 files before the problem is solved, you can contact with this email "hustzp@hust.edu.cn".



Huazhong University of Science and Technology