Electrocardiogram heartbeat classification based on a deep convolutional neural network and focal loss

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Electrocardiograms (ECGs) are crucial for diagnosing cardiovascular diseases and detecting arrhythmias.
• Traditional ECG analysis involves denoising, segmentation, feature extraction, and classification.
• Existing methods struggle with imbalanced datasets, particularly in classifying minority heartbeat categories.

Purpose of the Study:

• To develop an automated deep learning method for ECG analysis and arrhythmia classification.
• To address the challenge of imbalanced datasets in ECG classification.
• To improve the accuracy and performance of ECG-based arrhythmia detection.

Main Methods:

• A deep learning approach utilizing a convolutional neural network (CNN) for automatic feature extraction and classification.
• A novel heartbeat segmentation algorithm defining beats from R-peak to 1.2 times the median RR interval.
• Optimization of the CNN model with a focal loss function to enhance the importance of minority heartbeat classes.

Main Results:

• The proposed method achieved high performance metrics: 98.41% overall accuracy, 98.38% F1-score, 98.37% precision, and 98.41% recall.
• The focal loss function significantly improved classification accuracy for minority heartbeat classes.
• The method demonstrated superior performance compared to existing state-of-the-art techniques on the MIT-BIH and INCART datasets.

Conclusions:

• The proposed deep learning method offers an effective and automated approach for ECG analysis and arrhythmia detection.
• The focal loss function is a valuable tool for improving classification performance on imbalanced ECG datasets.
• This approach represents a significant advancement in automated cardiovascular disease diagnosis and arrhythmia monitoring.