Convolutional transformer-driven robust electrocardiogram signal denoising framework with adaptive parametric ReLU.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Signal Processing

Background:

• Electrocardiogram (ECG) is crucial for diagnosing cardiovascular diseases.
• ECG signals are susceptible to noise, hindering accurate clinical evaluation.
• Existing denoising methods may struggle with complex noise patterns and long time-series features.

Purpose of the Study:

• To develop an advanced deep learning framework for effective ECG signal denoising.
• To enhance the clinical utility of ECG recordings by reducing noise interference.
• To improve the accuracy of cardiovascular disease diagnosis through cleaner ECG data.

Main Methods:

• Proposed a novel Transformer-based convolutional neural network (APtrans-CNN) framework.
• Integrated transformers for global feature learning and CNNs for local feature learning.
• Introduced adaptively parametric ReLU (APReLU) and a dynamic feature aggregation module.

Main Results:

• APtrans-CNN accurately extracts pure ECG signals from noisy datasets.
• Demonstrated significant improvement in signal-to-noise ratio (SNR), increasing it from -4 dB to over 6 dB.
• Achieved diagnostic model accuracy exceeding 96% after denoising.

Conclusions:

• The APtrans-CNN framework offers effective ECG signal denoising capabilities.
• The model's ability to learn global and local features enhances its performance on complex time-series data.
• APtrans-CNN shows adaptability for various applications, improving cardiovascular diagnostics.