A Deep Neural Network Ensemble Classifier with Focal Loss for Automatic Arrhythmia Classification.

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence in Healthcare

Background:

• Automated electrocardiogram (ECG) classification is crucial for arrhythmia diagnosis, particularly for portable monitoring.
• Existing methods struggle with interpatient variability and accurate classification of supraventricular ectopic heartbeats (S).

Purpose of the Study:

• To develop an automated arrhythmia classification algorithm for improved interpatient assessment.
• To enhance the classification accuracy of supraventricular ectopic heartbeats (S) using a novel approach.

Main Methods:

• A new heartbeat segmentation method was developed to improve classification capacity.
• A combination of traditional sampling and focal loss was employed to address data imbalance.
• A deep convolutional neural network ensemble classifier was utilized within an interpatient evaluation paradigm.

Main Results:

• The algorithm achieved an overall accuracy of 91.89%, sensitivity of 85.37%, and specificity of 93.15%.
• For supraventricular ectopic heartbeats (S), the method demonstrated 80.23% sensitivity, 49.40% positivity, and 96.85% specificity.
• High classification performance was achieved in interpatient assessment without manual feature extraction or preprocessing.

Conclusions:

• The proposed automated algorithm offers a robust solution for arrhythmia classification, especially for supraventricular ectopic heartbeats (S).
• The interpatient evaluation paradigm and novel methods enhance the reliability of ECG analysis for portable devices.