CACHET-CADB

CACHET-CADB provides a context-rich ambulatory single-channel ECG database to support development and evaluation of arrhythmia detection algorithms, including atrial fibrillation detection for wearable devices under free-living conditions.

Key Features:

• Contextualized Data: Continuous contextual information recorded alongside ECG signals includes patient activities, body positions, movement accelerations, symptoms, stress levels, and sleep quality, with contextual information extracted every 10 seconds.
• Ambulatory Setting: The database comprises 259 days of single-channel ECG recordings from 24 patients collected in free-living ambulatory environments, with individual ECG records ranging from 24 hours to three weeks.
• Data Annotation and Quality: The resource contains 1,602 manually annotated 10-second heart-rhythm samples for algorithm training and evaluation.
• Noise Analysis: Approximately 11% of the ECG data are identified as noisy, reflecting real-world signal degradation in patient-operated wearable devices.
• Study Context: CACHET-CADB was developed as a component of the REAFEL study, which focuses on optimizing atrial fibrillation diagnosis among frail elderly patients.

Scientific Applications:

• Algorithm Development and Evaluation: The combined ECG and contextual dataset supports training and benchmarking of machine learning and deep learning arrhythmia detection algorithms under realistic ambulatory conditions.
• Improving Diagnostic Accuracy: Contextual variables enable refinement of signal interpretation to account for external factors and artifacts, aiming to improve reliability of arrhythmia detection in wearable devices.

Methodology:

Multi-site ambulatory data collection recorded continuous single-channel ECGs together with contextual data, with contextual information extracted every 10 seconds and manual annotation of 10-second heart-rhythm samples.