Deep learning-based electrocardiogram rhythm and beat features for heart abnormality classification.

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Electrocardiogram (ECG) signal classification is vital for diagnosing heart abnormalities.
• Deep learning (DL) offers potential for improved automated ECG analysis, but comprehensive evaluations are lacking.
• Previous studies often analyzed rhythm or beat features separately, limiting holistic assessment.

Purpose of the Study:

• To propose and evaluate a single DL architecture for classifying both ECG rhythm and beat patterns.
• To address the limitations of previous studies by analyzing rhythm and beat features concurrently.
• To enhance the accuracy and efficiency of automated ECG abnormality detection.

Main Methods:

• A deep learning (DL) model employing a one-dimensional convolutional neural network (1D-CNN) architecture was developed.
• The model was trained and validated on five databases encompassing nine rhythm and 15 beat classes.
• Experiments included varying datasets with different frequency samplings in intra- and inter-patient schemes.

Main Results:

• The DL model achieved high performance metrics for ECG rhythm classification, including 99.98% accuracy and 99.99% F1-score.
• For ECG beat classification, the model demonstrated strong results with 99.87% accuracy and 94.39% F1-score.
• The proposed architecture effectively discriminated between different ECG rhythm and beat assessments.

Conclusions:

• The study presents a robust DL methodology for comprehensive ECG analysis.
• The proposed 1D-CNN architecture offers an advanced tool for clinicians in detecting and discriminating heart abnormalities.
• This approach enhances the diagnostic capabilities of non-invasive ECG biomarkers.