Deep Time Growing Neural Network vs Convolutional Neural Network for Intelligent Phonocardiography

Area of Science:

• Artificial Intelligence
• Biomedical Engineering
• Cardiology

Background:

• Intelligent phonocardiography utilizes deep machine learning for heart disease diagnosis.
• Smart stethoscopes are being developed for decentralized cardiac condition assessment.
• Phonocardiography (PCG) signals are crucial for identifying cardiac abnormalities.

Purpose of the Study:

• To compare the performance of Deep Time Growing Neural Network (DTGNN) against Convolutional Neural Networks (CNN) for heart sound signal classification.
• To evaluate the effectiveness of DTGNN in discriminating between healthy individuals and patients with cardiac diseases using PCG data.
• To assess the adaptability of DTGNN in handling complexities in medical data, such as imbalanced training sets.

Main Methods:

• Utilized time series of heart sound signals (Phonocardiography - PCG) for analysis.
• Applied deep learning methods, specifically DTGNN and CNN, for signal classification.
• Employed the A-Test method to compare the structural risk between DTGNN and CNN.

Main Results:

• DTGNN demonstrated enhanced flexibility in learning subtle features within PCG signals.
• DTGNN exhibited superior capability in managing inherent complexities of medical data, including imbalanced training datasets.
• The study provides a comparative analysis of DTGNN and CNN performance on PCG data.

Conclusions:

• DTGNN presents a more flexible and robust deep learning approach for intelligent phonocardiography.
• The findings support the potential of DTGNN for developing advanced smart stethoscopes for improved cardiac disease diagnosis.
• DTGNN's ability to handle data complexities makes it a promising tool for real-world medical applications.