VitalDB Arrhythmia Database: An Anesthesiologist-Validated Large-scale Intraoperative Arrhythmia Dataset with Beat...

Area of Science:

• Cardiology
• Medical Informatics
• Anesthesiology

Background:

• Intraoperative cardiac arrhythmias have unique characteristics not well-represented in existing electrocardiogram (ECG) databases.
• Current ECG databases predominantly focus on ambulatory or intensive care settings, leaving a gap in surgical patient data.

Purpose of the Study:

• To introduce the VitalDB Arrhythmia Database, a novel, publicly accessible dataset of intraoperative ECG recordings.
• To provide a resource for developing and validating arrhythmia detection algorithms specifically for surgical patients.

Main Methods:

• Compilation of 734,528 seconds of continuous ECG data from 482 surgical patients.
• Development of a custom deep learning beat classifier for automated screening of arrhythmia segments.
• Rigorous annotation and validation of over 660,000 heartbeats across 10 rhythm categories by five anesthesiologists, ensuring high inter-rater reliability (Cohen's kappa = 0.930).

Main Results:

• The VitalDB Arrhythmia Database contains extensive, clinically validated intraoperative ECG data.
• High inter-rater reliability (Cohen's kappa = 0.930 ± 0.130) confirms the quality of annotations.
• A custom deep learning model was developed for efficient data processing and arrhythmia candidate identification.

Conclusions:

• The VitalDB Arrhythmia Database addresses a critical gap in research resources for intraoperative arrhythmias.
• This validated dataset will facilitate the development of advanced arrhythmia detection algorithms for surgical settings.
• The database enables further multimodal research into the hemodynamic effects of intraoperative arrhythmias.