Pattern discovery and similarity assessment for robust heart sound segmentation

Area of Science:

• Biomedical Engineering
• Signal Processing
• Cardiology

Background:

• Heart sound segmentation is crucial for Phonocardiogram (PCG) based pathology detection.
• Existing algorithms often lack robustness, performing poorly on diverse or low-quality datasets from varied acquisition systems.

Purpose of the Study:

• To develop and validate a robust heart sound segmentation algorithm for PCG signals.
• To integrate clinical information and pattern recognition for improved segmentation accuracy and generalization.

Main Methods:

• A pattern recognition approach was employed for segmenting fundamental heart sounds.
• The algorithm was integrated with clinical information.
• Validation was performed on a large, diverse database (3153 recordings, 764 patients) from multiple countries and acquisition systems.

Main Results:

• The algorithm achieved high sensitivity (95%) and positive predictivity (96%).
• Demonstrated robust performance across varied databases and acquisition instrumentations.
• Maintained generalization capabilities despite data diversity.

Conclusions:

• The proposed segmentation algorithm is accurate and robust for PCG signal analysis.
• It shows potential for reliable pathology detection in real-world clinical settings.
• The integration of clinical information enhances segmentation performance and generalizability.