VitalDB Arrhythmia Database: An Anesthesiologist-Validated Large- scale Intraoperative Arrhythmia Dataset with Beat and Rhythm Labels

Comprehensive Summary

The VitalDB Arrhythmia Database serves as a publicly accessible database containing labeled, intraoperative ECG recordings that can serve as source to develop further arrhythmia detection tools. Navigating the extensive source data for the creation of the VitalDB database involved using the UniMS-ECGNet deep learning algorithm to classify ECG beats and identify arrhythmia segments. Particular features of importance included beat annotation, rhythm annotation, signal quality assessment, transient conduction block events, patterned arrhythmias, and sinus node dysfunction. All segments and annotations presented by the deep learning algorithm were further reviewed and validated by five anesthesiologists prior to incorporation into the database.

Outcomes and Implications

The VitalDB Arrhythmia Database makes intraoperative arrhythmia data available to the scientific community, which paves the way for further research into improving intraoperative patient care. The necessity of physician review and validation in the process of producing the VitalDB database shows the continuing need for human review when using artificial intelligence (AI) based-models in the clinical research space. As with the VitalDB database, other databases can be created using the assistance of AI-driven learning models, which increases the usability of the extensive stores of data that we possess.