Parralel Recurrent Convolutional Neural Network for Abnormal Heart Sound Classification

Area of Science:

• Cardiology
• Artificial Intelligence
• Signal Processing

Background:

• Heart abnormalities pose a significant global health challenge.
• Accurate and early detection of cardiac conditions is crucial for effective treatment.
• Automated analysis of heart sounds offers a non-invasive diagnostic approach.

Purpose of the Study:

• To introduce and evaluate a Parallel Convolutional Neural Network (PCNN) for detecting heart abnormalities using heart sound signals.
• To compare the performance of PCNN against other deep learning models.
• To assess the potential of PCNN for real-world clinical decision support.

Main Methods:

• A Parallel Convolutional Neural Network (PCNN) architecture was developed, combining recurrent and convolutional neural networks.
• The PCNN was trained and tested on the public Physionet heart sound dataset.
• Performance was benchmarked against Serial Convolutional Neural Network (SCNN), Long- and Short-Term Memory (LSTM) networks, and Conventional CNN (CCNN).

Main Results:

• The PCNN achieved an accuracy of 87.2% in detecting heart abnormalities.
• PCNN outperformed SCNN by 12%, LSTM by 7%, and CCNN by 0.5%.
• The model demonstrated superior performance in preserving dynamic signal content.

Conclusions:

• The PCNN model shows significant promise for accurate heart abnormality detection.
• Its high accuracy and efficiency make it suitable for integration into Internet of Things (IoT) platforms.
• PCNN can serve as a valuable decision support tool for cardiac screening.