Heart sound classification based on equal scale frequency cepstral coefficients and deep learning.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Cardiology

Background:

• Heart diseases are a leading cause of mortality, necessitating early detection and prevention strategies.
• Mel-scale Frequency Cepstral Coefficients (MFCC) are commonly used for heart sound analysis but may not be optimal due to their basis in human auditory properties.
• The frequency characteristics of heart sounds differ from those processed by the human auditory system.

Purpose of the Study:

• To introduce a novel feature extraction method, Equal-scale Frequency Cepstral Coefficients (EFCC), for improved heart sound analysis.
• To develop and evaluate advanced classifiers integrating Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), and Random Forest (RF) for heart sound classification.
• To compare the performance of EFCC with MFCC in identifying cardiac abnormalities.

Main Methods:

• Proposed EFCC feature by replacing Mel-scale filters with equally spaced triangular overlapping filters.
• Developed hybrid CNN-RNN-RF classifiers to capture both spatial and temporal features of heart sounds.
• Validated the algorithm using a private database and the PhysioNet CinC 2016 Challenge Database.

Main Results:

• EFCC features demonstrated superior performance and robustness compared to MFCC features in classifying heart sounds from new patients.
• Ten-fold cross-validation confirmed the effectiveness of the proposed EFCC-based approach.
• The developed algorithm achieved high precision in heart sound classification.

Conclusions:

• EFCC offers a more objective and suitable feature for heart sound signal processing than MFCC.
• The hybrid CNN-RNN-RF classifier effectively extracts relevant information for accurate heart sound classification.
• This research paves the way for advanced, real-time heart monitoring using wearable medical devices.