CACHET-CADB: A Contextualized Ambulatory Electrocardiography Arrhythmia Dataset

Area of Science:

• Biomedical Engineering
• Cardiology
• Data Science

Background:

• Wearable electrocardiogram (ECG) devices are increasingly used for ambulatory arrhythmia monitoring.
• Existing arrhythmia detection algorithms struggle with real-world data due to signal quality differences compared to clean clinical databases.
• Patient-operated wearable ECGs present unique challenges in signal quality and artifact presence.

Purpose of the Study:

• To design and develop the CACHET-CADB, a multi-site, contextualized ECG database for evaluating arrhythmia detection algorithms in free-living conditions.
• To address the limitations of current public ECG databases that do not represent ambulatory recordings from patient-operated devices.
• To facilitate the development of robust arrhythmia detection algorithms for wearable ECGs by incorporating contextual data.

Main Methods:

• The CACHET-CADB database was created as part of the REAFEL study, focusing on frail elderly patients for atrial fibrillation diagnosis.
• It includes continuous ECG recordings alongside contextual data: activities, body positions, movement accelerations, symptoms, stress, and sleep quality.
• Data was collected from 24 patients over 259 days, yielding 1,602 annotated 10-second heart rhythm samples, with ECG record lengths from 24 hours to 3 weeks.

Main Results:

• Nearly 11% of the ECG data within the CACHET-CADB was identified as noisy.
• Contextual data was extracted for every 10-second interval, providing detailed ambulatory information.
• The database provides a valuable resource for training and validating machine/deep learning models for arrhythmia detection.

Conclusions:

• The CACHET-CADB database offers a significant advancement for developing arrhythmia detection algorithms for wearable devices in real-world settings.
• Incorporating contextual data alongside ECG signals is crucial for improving algorithm accuracy and robustness.
• The availability of this database and its associated software toolkit will aid researchers in advancing ambulatory arrhythmia monitoring.