An open access database for the evaluation of heart sound algorithms

Area of Science:

• Biomedical Engineering
• Cardiology
• Digital Health

Background:

• Automated phonocardiogram (PCG) analysis for heart sound segmentation and classification shows potential for accurate clinical pathology detection.
• Comparative algorithm analyses are limited by the absence of standardized, high-quality, and validated open heart sound databases.

Purpose of the Study:

• To introduce a public heart sound database for the PhysioNet/Computing in Cardiology (CinC) Challenge 2016.
• To facilitate rigorous validation and standardization of heart sound analysis algorithms.

Main Methods:

• Assembled a comprehensive public heart sound database comprising nine distinct datasets from global research groups.
• Included 2435 heart sound recordings from 1297 healthy subjects and patients with various cardiac conditions.
• Detailed subject demographics, recording parameters (environment, equipment, duration), and synchronously recorded signals.

Main Results:

• The database contains diverse recordings from varied clinical and non-clinical settings.
• Provides detailed metadata including subject demographics, recording specifics, and sensor information.
• Includes commonly used segmentation and classification methods with open-source code.

Conclusions:

• The PhysioNet/CinC Challenge 2016 database addresses the need for standardized, high-quality heart sound data.
• Facilitates improved comparative analysis and development of automated PCG analysis algorithms.
• Promotes advancements in clinical applications for heart disease detection through accessible data and tools.