PTB-XL ECG

心电图 (ECG) 是评估患者心脏状况的关键诊断工具。自动ECG解释算法作为诊断支持系统，仅根据常规服用的ECG数量，可为医务人员带来巨大的缓解。但是，此类算法的开发需要大量的训练数据集和清晰的基准程序。我们认为，现有的可免费访问的ECG数据集不能令人满意地涵盖这两个方面。
Ptb-xl ECG数据集是来自10秒长度的18885名患者的21837个临床12导联ECG的大型数据集。原始波形数据由最多两名心脏病专家注释，他们为每个记录分配了潜在的多个ECG声明。总共71种不同的ECG语句符合scp-ecg标准，涵盖了诊断，形式和节律语句。为了确保在数据集上训练的机器学习算法的可比性，我们提供了建议的分为训练集和测试集。结合广泛的注释，这将数据集变成了用于训练和评估自动ECG解释算法的丰富资源。数据集由人口统计学，梗塞特征，诊断性ECG语句的可能性以及注释的信号属性的广泛元数据补充。

Electrocardiography (ECG) is a critical diagnostic tool for evaluating a patient’s cardiac condition. Automatic ECG interpretation algorithms, as diagnostic support systems, can greatly alleviate the workload of medical personnel based solely on the volume of routinely recorded ECGs. However, the development of such algorithms demands large-scale training datasets and well-defined benchmark protocols. We argue that existing freely accessible ECG datasets fail to satisfactorily cover both of these aspects.

The PTB-XL ECG dataset is a large-scale collection of 21,837 clinical 12-lead ECG recordings from 18,885 patients, each with a 10-second duration. The raw waveform data were annotated by up to two cardiologists, who assigned multiple plausible ECG statements to each recording. In total, 71 distinct ECG statements adhering to the SCP-ECG standard are included, spanning diagnostic, morphological, and rhythmic categories. To ensure the comparability of machine learning algorithms trained on this dataset, we provide a recommended train-test split. Coupled with the extensive annotations, this transforms the dataset into a rich resource for training and evaluating automatic ECG interpretation algorithms. The dataset is supplemented with comprehensive metadata including demographic information, infarction characteristics, the likelihood of diagnostic ECG statements, and annotated signal attributes.

搜集汇总

背景与挑战

背景概述

PTB-XL ECG是一个大型临床心电图数据集，包含来自18,885名患者的21,837个12导联ECG记录，每个记录长度为10秒，并由心脏病专家标注了71种符合scp-ecg标准的ECG语句。该数据集提供了训练集和测试集的划分，并附有丰富的元数据，专为训练和评估自动ECG解释算法设计，旨在解决现有ECG数据在算法开发中的不足。