PTB-XL+, a comprehensive electrocardiographic feature dataset.

Area of Science:

• Biomedical Engineering
• Cardiology
• Artificial Intelligence

Background:

• Machine learning (ML) analysis of electrocardiography (ECG) data is growing, aided by large public datasets.
• Current datasets lack essential ECG features crucial for cardiologists and automatic analysis algorithms.
• These vital ECG features are often proprietary and inaccessible.

Purpose of the Study:

• To enrich the PTB-XL dataset with derived ECG features and automatic diagnostic statements.
• To enable direct comparison of ML models trained on clinical versus automatically generated labels.
• To enhance the PTB-XL dataset's utility as a reference for ML in ECG analysis.

Main Methods:

• Incorporated ECG features from two leading commercial algorithms and an open-source implementation.
• Added automatic diagnostic statements from commercial ECG analysis software.
• Performed extensive technical validation of the added features and diagnostic statements for ML applications.

Main Results:

• Successfully integrated proprietary and open-source ECG features into the PTB-XL dataset.
• Included a comprehensive set of automatic diagnostic statements.
• Validated the technical accuracy and usability of these additions for ML.

Conclusions:

• The enhanced PTB-XL dataset provides previously inaccessible, clinically relevant ECG features.
• This resource facilitates the development and comparison of ML models for ECG analysis.
• The release significantly boosts the PTB-XL dataset's value as a benchmark for ML in cardiology.