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"

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".本测试数据集包含800份24小时动态心电图（Holter）记录，均来自论文中随机临床队列的测试子集。每份记录的RR间期数据均提取自华中科技大学同济医院心内科心脏功能检查中心下属3个院区（主院区、光谷院区、中法新城院区）使用DMS Holter公司（美国内华达州Stateline市）生产的Holter记录仪采集的24小时12导联动态心电图。所有数据首先由初级心内科医师进行初步判读，随后经3名持有执业资质的资深心内科医师复核，以确保基础诊断标签的准确性。本次心内科医师团队对标注后的记录进行共识讨论，并为模型评估提供了参考标准。对于阵发性心房颤动（Paroxysmal Atrial Fibrillation, PAF）患者，每一次心房颤动（Atrial Fibrillation, AF）发作均被精准标注了起始与终止时间：PAF发作的起始时间为心房率＞350次/分时的首个心房波对应时刻，终止时间为AF终止后首个窦性心律P波对应的时刻。具体而言，持续时长超过30秒的PAF发作将被精准标注，时长不足30秒的PAF发作则尽可能精准标注。此外，每一段早搏或心动过速间期均被标记为"A"（心房事件）或"V"（心室事件）；因QRS波脱落导致的长RR间期则被标记为"B"，用于进一步标注二度房室传导阻滞。本800份记录中包含200份全程心房颤动（Whole-course Atrial Fibrillation, WAF）记录、200份PAF记录与400份非心房颤动（Non-Atrial Fibrillation, NAF）记录。全程AF患者的完整记录仅包含AF心电信号；非AF患者的完整记录则不含AF心电信号，但包含窦性心律、窦性心律不齐、房性心律失常、室性心律失常、房室传导阻滞等心电信号；PAF患者的数据则同时包含AF发作间期与非AF心电信号。全程AF患者与PAF患者均被归类为AF患者。每份记录的类型由文件名标识，数据内容的相关注释可参见图片"Note.jpg"。