A deep convolutional neural network model to classify heartbeats.

Area of Science:

• Cardiology
• Artificial Intelligence
• Signal Processing

Background:

• Electrocardiogram (ECG) is crucial for monitoring heart activity and diagnosing cardiac abnormalities like arrhythmia.
• Arrhythmia diagnosis relies on classifying individual heartbeats based on ECG morphology, a process complicated by signal noise.
• Heartbeats are categorized into five types: non-ectopic, supraventricular ectopic, ventricular ectopic, fusion, and unknown.

Purpose of the Study:

• To develop and evaluate a deep convolutional neural network (CNN) for automatic identification and classification of five distinct heartbeat categories in ECG signals.
• To assess the CNN's performance on both original and noise-attenuated ECG datasets.

Main Methods:

• A 9-layer deep convolutional neural network (CNN) was designed for heartbeat classification.
• ECG data from a public database was used, with artificial augmentation to balance the 5 heartbeat classes and filtering to reduce high-frequency noise.
• The CNN was trained on augmented data and evaluated on both noisy and noise-free ECG signals.

Main Results:

• The CNN achieved high accuracy in classifying heartbeats: 94.03% on original ECGs and 93.47% on noise-free ECGs when trained with balanced, augmented data.
• Training with imbalanced data resulted in lower accuracy (89.07% on noisy, 89.3% on noise-free ECGs).

Conclusions:

• A properly trained CNN model demonstrates significant potential as an effective tool for ECG screening.
• The model can rapidly identify various types and frequencies of arrhythmic heartbeats, aiding clinical diagnosis.