Real-world 24h+ ECG dataset with quality annotations and motion context

Area of Science:

• Biomedical Engineering
• Medical Informatics
• Cardiology

Background:

• Accurate electrocardiogram (ECG) signal quality assessment is crucial for reliable analysis, particularly in long-term, free-living recordings.
• The absence of publicly available, annotated datasets impedes the development and benchmarking of automated ECG quality assessment algorithms.

Purpose of the Study:

• To introduce the BUT ECG Quality Database (BUT QDB), a novel, freely accessible dataset for ECG signal quality assessment.
• To provide a standardized evaluation protocol and open-source code to facilitate objective benchmarking of automated algorithms.

Main Methods:

• Compilation of 18 long-term, single-lead ECG and 3-axis accelerometer (ACC) recordings from 15 healthy subjects, totaling over 86 hours.
• Expert manual annotation of ECG signals, computationally expanded to sample-by-sample labels for approximately 2.2 million cardiac cycles.
• Categorization of signal quality into three distinct levels, with ACC data enabling motion-informed or ACC-only assessment.

Main Results:

• Creation of the BUT QDB, a comprehensive dataset with expert-verified ECG signal quality labels.
• Inclusion of ACC data for multimodal signal quality analysis.
• Development of a standardized protocol and open-source tools for algorithm evaluation.

Conclusions:

• The BUT QDB addresses the critical need for annotated data in ECG signal quality assessment research.
• This resource will enable objective benchmarking and enhance the comparability of diverse automated ECG quality assessment methods.
• Facilitation of advancements in reliable, long-term ECG monitoring and analysis.