An ECG Signal Classification Method Based on Dilated Causal Convolution.

Area of Science:

• Cardiology and Medical Informatics
• Artificial Intelligence in Healthcare
• Signal Processing and Machine Learning

Background:

• Rising incidence and younger trends in cardiovascular disease necessitate efficient diagnostic tools.
• Existing medical resources are strained, highlighting the need for automated medical signal analysis.
• Limitations of recurrent neural networks in hardware acceleration for real-time ECG analysis.

Purpose of the Study:

• To propose an automated ECG signal classification method using a dilated causal convolutional neural network (CNN).
• To overcome hardware acceleration limitations associated with recurrent neural networks.
• To improve the accuracy and efficiency of cardiovascular disease detection through advanced deep learning.

Main Methods:

• Development of a dilated causal CNN architecture incorporating fully convolutional networks and causal convolution.
• Integration of dilated factors to manage network depth and mitigate gradient issues (explosion/disappearance).
• Inclusion of residual blocks with shortcut connections to enhance model performance and stability.

Main Results:

• The proposed dilated causal CNN model demonstrated effectiveness in ECG signal classification.
• Validation using the MIT-BIH Atrial Fibrillation Database (MIT-BIH AFDB) yielded a classification accuracy of 98.65%.
• The architecture successfully addressed the computational constraints of traditional recurrent networks.

Conclusions:

• The dilated causal CNN offers a robust and accurate solution for automated ECG classification.
• This approach provides a viable alternative to recurrent networks, enabling better hardware acceleration.
• The high accuracy achieved suggests significant potential for clinical application in cardiovascular diagnostics.