Towards the classification of heart sounds based on convolutional deep neural network

Area of Science:

• Biomedical Engineering
• Cardiology
• Artificial Intelligence in Healthcare

Background:

• Heart sounds contain crucial diagnostic information for early detection of cardiac diseases.
• Existing methods for heart disease detection often rely on various signal-processing techniques applied to heart sounds.

Purpose of the Study:

• To introduce an advanced methodology for heart disease detection based on the analysis of heart sounds.
• To leverage deep learning for enhanced accuracy in diagnosing heart conditions from auditory signals.

Main Methods:

• The proposed methodology involves three stages: spectrogram generation, deep feature extraction, and classification.
• Heart sounds are transformed into spectrogram images using time-frequency transformation.
• Deep features are extracted using pre-trained convolutional neural network models (AlexNet, VGG16, VGG19), followed by Support Vector Machine classification.

Main Results:

• The method was evaluated on two datasets from The Classifying Heart Sounds Challenge.
• Deep features were successfully extracted using established CNN architectures.
• The Support Vector Machine classifier was employed for the final diagnostic stage.

Conclusions:

• The proposed heart sound analysis method demonstrated superior performance compared to existing approaches.
• The integration of spectrograms, deep feature extraction, and SVM classification offers a promising avenue for improved heart disease detection.