A comparative analysis of Constant-Q Transform, gammatonegram, and Mel-spectrogram techniques for AI-aided cardiac diagnostics

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Cardiology

Background:

• Cardiovascular diseases (CVDs) are a leading cause of global mortality, necessitating early detection of cardiac abnormalities.
• Abnormal heart sounds, crucial indicators of cardiac issues, are challenging to identify due to their low-frequency characteristics.

Purpose of the Study:

• To develop and evaluate an automated system for heart sound classification using deep learning.
• To enhance the early and accurate detection of cardiac abnormalities through Phonocardiogram (PCG) analysis.

Main Methods:

• Employed digital signal processing and advanced deep learning techniques for automated heart sound classification.
• Utilized time-frequency representations: gammatonegram, Mel-spectrogram, and Constant-Q Transform (CQT).
• Designed and evaluated a Convolutional Neural Network with Directed Acyclic Graph (DAG-CNN) architecture.

Main Results:

• Achieved high classification accuracies: 100% for gammatonegram, 99.7% for Mel-spectrogram, and 99.5% for CQT.
• Demonstrated superior performance compared to pre-trained Convolutional Neural Network (CNN) models.
• The proposed DAG-CNN model effectively classifies heart sounds using various time-frequency representations.

Conclusions:

• The developed automated heart sound classification system shows significant promise for early cardiac abnormality detection.
• This advancement offers a cost-effective solution, particularly beneficial for resource-limited settings.
• The study contributes to bridging the diagnostic gap and improving cardiac care accessibility.