ECG Signal Classification Using Deep Learning Techniques Based on the PTB-XL Dataset

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Electrocardiogram (ECG) signal analysis is crucial for diagnosing cardiovascular diseases.
• Machine learning algorithms are increasingly utilized for ECG signal classification.
• Accurate and efficient ECG analysis is vital for timely disease diagnosis and patient management.

Purpose of the Study:

• To develop and evaluate deep neural network architectures for automatic classification of primary ECG signals.
• To compare the performance of different network architectures, including convolutional networks and SincNet, with and without entropy-based features.
• To assess the classification accuracy and computational efficiency of the proposed models across varying numbers of disease classes.

Main Methods:

• Development of three deep neural network architectures: a standard convolutional network, a SincNet, and a convolutional network augmented with entropy-based features.
• Utilized the PTB-XL database for training and testing the models.
• Dataset split into training (70%), validation (15%), and test (15%) sets.
• Evaluated model performance for 2, 5, and 20 disease classes.

Main Results:

• The convolutional network incorporating entropy-based features achieved the highest classification accuracy.
• The standard convolutional network (without entropy features) demonstrated slightly lower accuracy but superior computational efficiency due to fewer neurons.
• Performance was evaluated across 2, 5, and 20 disease classes, indicating robustness of the models.

Conclusions:

• Deep neural networks, particularly convolutional networks with entropy features, show significant promise for accurate ECG signal classification in cardiovascular disease diagnosis.
• The trade-off between classification accuracy and computational efficiency can be managed by architectural choices, such as the inclusion of entropy features.
• The developed models offer a viable approach for automated ECG analysis, potentially improving diagnostic workflows.