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

Reliable assessment of electrocardiogram (ECG) signal quality is critical for ensuring accurate analysis, especially in long-term recordings acquired in free-living conditions. However, the development and benchmarking of automatic quality assessment algorithms are hindered by the lack of publicly available, annotated datasets. To address this, we introduce the BUT ECG Quality Database (BUT QDB) - a freely available dataset containing 18 long-term recordings (single-lead ECG and 3-axis ACC) acquired from 15 healthy subjects. The dataset includes more than 86 hours of expert-labeled ECG signals. Primary manual annotations were performed on a segment basis by three independent ECG experts. These were then computationally expanded to sample-by-sample labels representing the underlying 2.2 million cardiac cycles. The quality is categorized into three levels. ACC data allow the investigation of motion-informed or ACC-only signal quality assessment. In addition, we provide a standardized evaluation protocol and open-source code designed for benchmarking automated ECG quality assessment algorithms. BUT QDB aims to enable objective benchmarking and improve the comparability of results across different ECG signal quality assessment methods.