PTB-XL, a large publicly available electrocardiography dataset

Area of Science:

• Cardiology
• Biomedical Engineering
• Data Science

Background:

• Electrocardiography (ECG) is crucial for diagnosing cardiovascular diseases.
• Machine learning algorithms are increasingly used for ECG interpretation.
• Lack of public datasets and standardized benchmarking hinders algorithm development.

Purpose of the Study:

• To introduce PTB-XL, the largest freely accessible clinical 12-lead ECG dataset.
• To provide a resource for developing and evaluating automated ECG interpretation algorithms.
• To address the limitations of existing ECG datasets and benchmarking procedures.

Main Methods:

• Compiled 21,837 ECG records (10-second length) from 18,885 patients.
• Annotated ECG data with multi-label diagnostic classifications by cardiologists.
• Aggregated diagnostic labels into super and subclasses.
• Included metadata: demographics, diagnostic statements, likelihoods, signal properties, and suggested data splits.

Main Results:

• PTB-XL is the largest publicly available clinical 12-lead ECG dataset to date.
• The dataset includes a wide range of diagnostic classes, with a significant proportion of healthy records.
• Rich metadata facilitates comprehensive algorithm development and evaluation.

Conclusions:

• PTB-XL serves as a valuable resource for advancing machine learning in ECG analysis.
• The dataset enables standardized benchmarking for automated ECG interpretation algorithms.
• Facilitates research in cardiovascular disease diagnosis using AI.