Machine Learning based Heart Murmur Detection and Classification.

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Cardiovascular diseases are a leading global cause of death.
• Early detection of heart conditions, like murmurs, is crucial for effective treatment.
• Phonocardiogram (PCG) signals offer valuable diagnostic information.

Purpose of the Study:

• To develop a novel machine learning (ML) framework for early detection and classification of heart murmurs.
• To analyze heart murmur characteristics including presence, clinical outcome, and quality.
• To leverage ML for improved cardiovascular disease diagnosis using PCG signals.

Main Methods:

• A multi-stage ML pipeline was designed for heart murmur analysis.
• Transfer learning methods were employed for initial murmur presence classification.
• 1D convolution and audio spectrogram transformers were used for clinical outcome classification.
• Wav2Vec encoder and AdaBoost classifier were utilized for murmur quality identification.

Main Results:

• Achieved 81.08% validation accuracy for murmur presence classification.
• Obtained 68.23% validation accuracy for clinical outcome classification.
• Demonstrated 60.52% sensitivity and 74.46% specificity in classification tasks.
• Utilized the PhysioNet 2022 dataset for model training and validation.

Conclusions:

• The proposed ML framework shows promise for PCG-based heart murmur detection and classification.
• This approach facilitates early identification and quality analysis of heart murmurs.
• The findings have significant implications for the diagnosis and management of cardiovascular diseases.