Inter-patient arrhythmia classification with improved deep residual convolutional neural network.

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Early detection of arrhythmias is crucial for reducing cardiovascular disease mortality.
• Electrocardiogram (ECG) analysis is vital for diagnosing arrhythmias.
• Segment-based ECG classification is preferred for clinical settings.

Purpose of the Study:

• To develop an improved deep residual convolutional neural network for automated arrhythmia classification from ECG segments.
• To enhance the clinical applicability of ECG-based arrhythmia detection.

Main Methods:

• Utilized overlapping segmentation to create 5-second ECG segments from the MIT-BIH database, addressing class imbalance.
• Applied discrete wavelet transform (DWT) for denoising ECG segments.
• Employed an improved deep residual convolutional neural network with focal loss for classification.

Main Results:

• Achieved high performance for normal segments (94.54% sensitivity, 93.33% positive predictivity, 80.80% specificity).
• Demonstrated strong results for ventricular ectopic segments (88.35% sensitivity, 79.86% positive predictivity, 94.92% specificity).
• Showcased improved classification performance with the focal loss function.

Conclusions:

• The proposed deep residual convolutional neural network model demonstrates comparable performance to existing methods.
• Overlapping segmentation and focal loss significantly improve arrhythmia classification accuracy.
• The method shows promise for clinical application in automated arrhythmia detection.