Deep Learning Techniques in the Classification of ECG Signals Using R-Peak Detection Based on the PTB-XL Dataset

Area of Science:

• Computational intelligence
• Biomedical signal processing
• Machine learning for healthcare

Background:

• Deep Neural Networks (DNNs) show promise for analyzing electrocardiographic (ECG) signals.
• Limited research exists on optimizing DNNs for ECG data analysis and classification.
• Effective ECG recognition requires exploring various signal information combinations.

Purpose of the Study:

• To investigate the effectiveness of combining raw ECG signals, entropy-based features, and QRS complexes for heart disease classification using DNNs.
• To determine the optimal combination of signal information for improved ECG recognition accuracy.
• To present an innovative method for multi-lead QRS complex extraction.

Main Methods:

• A convolutional neural network (CNN) was developed for encoding single QRS complexes.
• Entropy-based features were computed from raw ECG signals and extracted QRS complexes.
• An innovative QRS complex extraction method aggregated results from established algorithms using k-mean clustering for multi-lead signals.

Main Results:

• The combination of raw ECG signals with entropy-based features and extracted QRS complexes yielded the best classification results.
• Using raw signals with entropy-based features but without extracted QRS complexes resulted in significantly poorer performance.
• The proposed QRS complex extraction method was successfully integrated into the DNN framework.

Conclusions:

• Incorporating entropy-based features and extracted QRS complexes alongside raw ECG data is beneficial for DNN-based heart disease classification.
• The study highlights the importance of feature engineering and signal segmentation for enhancing ECG analysis.
• The findings suggest a promising direction for developing more accurate automated diagnostic tools for cardiovascular conditions.