Cardi-Net: A deep neural network for classification of cardiac disease using phonocardiogram signal

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Signal Processing

Background:

• Limited medical facilities in isolated regions hinder timely cardiovascular disease diagnosis, contributing to high mortality rates.
• Phonocardiogram (PCG) signals offer valuable diagnostic information for cardiac conditions.

Purpose of the Study:

• To develop an automatic deep learning system for diagnosing multiple cardiac diseases from PCG signals.
• To address the challenge of delayed diagnosis in remote areas through accessible technology.

Main Methods:

• A deep learning model combining a convolutional neural network (CNN) and power spectrogram Cardi-Net was employed.
• Power Spectral Density (PSD) analysis was utilized to extract discriminatory features from PCG signals for multi-classification.
• Data augmentation techniques were applied to enhance model robustness.

Main Results:

• The model achieved an overall accuracy of 98.879% in diagnosing multiple heart diseases on the test dataset.
• The system demonstrated high reliability and robustness through 10-fold cross-validation.
• Power spectrogram conversion from PCG signals was rapid, ranging from 0.10s to 0.11s.

Conclusions:

• The proposed model offers a completely automatic solution, eliminating the need for signal pre-processing and feature engineering.
• The system's low complexity and fast processing time make it suitable for real-time applications.
• The architecture's deployability on various platforms (cloud, low-cost processors, mobile apps) ensures accessibility in remote dispensaries.