Automatic QRS complex detection using two-level convolutional neural network.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• QRS complex detection is crucial for electrocardiogram (ECG) analysis.
• Existing methods often rely on complex manual features, limiting adaptability.
• Fixed parameters in traditional methods struggle with diverse QRS morphologies.

Purpose of the Study:

• To develop an accurate and automatic QRS complex detection method.
• To overcome the limitations of traditional hand-crafted feature-based approaches.
• To improve the computational efficiency and robustness of QRS detection.

Main Methods:

• Utilized a 1-D convolutional neural network (CNN) architecture.
• Employed object-level and part-level CNNs for automatic feature extraction.
• Integrated a simple temporal domain difference-based ECG signal preprocessing technique.
• Used a multi-layer perceptron (MLP) for final QRS complex detection.

Main Results:

• Achieved high performance on the MIT-BIH arrhythmia database.
• Reported sensitivity (Sen) of 99.77% and positive predictivity rate (PPR) of 99.91%.
• Demonstrated a low detection error rate (DER) of 0.32% across varying signal-to-noise ratios (SNR).

Conclusions:

• Proposed an automatic QRS detection method leveraging a two-level 1-D CNN.
• The method offers comparable accuracy to state-of-the-art approaches.
• The approach provides an efficient and robust solution for ECG analysis.