Digital auscultation analysis for heart murmur detection.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Cardiology

Background:

• Heart murmurs, often caused by valve disorders, require accurate detection for diagnosis.
• Phonocardiographic signals are complex and variable, posing challenges for automated analysis.
• Existing murmur detection methods vary in their feature extraction and classification approaches.

Purpose of the Study:

• To compare the effectiveness of different feature families for murmur detection in phonocardiographic signals.
• To evaluate the performance of time-varying/time-frequency, perceptual, and fractal features.
• To identify the most robust and accurate feature set for automated murmur detection.

Main Methods:

• Analysis of three feature families: time-varying/time-frequency, perceptual, and fractal.
• Systematic testing of feature combinations to improve accuracy.
• Evaluation using a k-nearest neighbors classifier on a database of 164 phonocardiographic recordings.
• Extraction of 360 individual heartbeats (180 normal, 180 with murmurs).

Main Results:

• Fractal features achieved the highest accuracy (97.17%), followed by time-varying/time-frequency (95.28%), and perceptual features (88.7%).
• A simplified approach using only two fractal features yielded approximately 94% accuracy.
• Fractal features demonstrated superior robustness in terms of accuracy versus computational load.

Conclusions:

• Fractal features are the most robust and computationally efficient parameters for automated murmur detection.
• The fractal feature-based approach offers a viable alternative, especially considering the complexities of automated segmentation for other feature types.
• This study highlights the potential of fractal analysis in improving diagnostic accuracy for cardiac auscultation signals.