Automatic heart sound segmentation and murmur detection in pediatric phonocardiograms

Area of Science:

• Cardiology
• Biomedical Engineering
• Digital Signal Processing

Background:

• Digital analysis of heart sounds is a growing research area.
• Previous attempts to create decision support systems for heart sound analysis have been made.
• Accurate segmentation and murmur detection are crucial for diagnosing cardiac conditions.

Purpose of the Study:

• To propose two novel algorithms for heart sound analysis.
• To develop an algorithm for segmenting heart sounds into cardiac cycles.
• To develop an algorithm for detecting heart murmurs from phonocardiogram (PCG) signals.

Main Methods:

• Heart sound segmentation using the autocorrelation function to identify periodic components in PCG signals.
• Heart murmur detection utilizing features from multiple domains.
• Algorithm evaluation through random train/test set division and patient-specific data splitting.

Main Results:

• The segmentation algorithm achieved 89.2% sensitivity and 98.6% positive predictive value.
• Murmur detection with random data division yielded 98.42% sensitivity and 97.21% specificity (2.19% minimum error).
• Patient-specific data division for murmur detection showed significantly worse performance (33.65% minimum error).

Conclusions:

• The proposed segmentation algorithm demonstrates high accuracy in identifying heart cycles.
• The murmur detection algorithm shows promise but requires further refinement for robust clinical application, especially concerning patient variability.
• Further research is needed to improve the generalization of murmur detection algorithms across different patient cohorts.