[Research on electrocardiogram classification using deep residual network with pyramid convolution structure].

Area of Science:

• Cardiology
• Artificial Intelligence
• Signal Processing

Background:

• Deep neural networks (DNNs) are increasingly used for electrocardiogram (ECG) signal classification.
• Existing DNN models often struggle with effective feature extraction from raw ECG data.

Purpose of the Study:

• To propose a novel deep residual network model, PC-DRN, for improved ECG signal classification.
• To enhance feature extraction capabilities for more accurate arrhythmia detection.

Main Methods:

• Developed a deep residual network model incorporating pyramidal convolutional (PC) layers.
• PC layers were designed to extract multi-scale features from raw ECG signals simultaneously.
• Trained and validated the model on the PhysioNet Computing in Cardiology Challenge 2017 (CinC2017) dataset for 4 types of ECG data.

Main Results:

• The PC-DRN model demonstrated significant improvements in ECG classification performance.
• Average sequence-level F1 (SeqF1) score increased from 0.857 to 0.920.
• Average set-level F1 (SetF1) score improved from 0.876 to 0.925.

Conclusions:

• The proposed PC-DRN model offers a promising approach for ECG signal feature extraction and classification.
• This model provides an effective tool for accurate arrhythmia classification, outperforming previous methods.