An Effective and Lightweight Deep Electrocardiography Arrhythmia Recognition Model Using Novel Special and Native...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Deep learning models are increasingly used for electrocardiography (ECG) analysis in clinical diagnosis.
• Lightweight and effective deep learning models are crucial for real-world deployment in cardiac arrhythmia detection.

Purpose of the Study:

• To propose two effective and lightweight deep learning models, Deep-SR and Deep-NSR, for accurate ECG beat recognition.
• To evaluate the performance and generalization of these models compared to state-of-the-art methods.

Main Methods:

• ECG beats from the MIT-BIH dataset were transformed into 2D RGB images for input into 2D CNN models.
• Models were optimized using layer initialization, on-the-fly augmentation, regularization, Adam optimizer, and weighted random sampler.
• Performance was assessed via stratified 5-fold cross-validation.

Main Results:

• The Deep-SR and Deep-NSR models achieved high overall accuracy (99.93% and 99.96%) in recognizing normal beats and specific arrhythmias (ventricular ectopic, supraventricular ectopic, fusion).
• The proposed models demonstrated superior effectiveness and generalization compared to existing state-of-the-art models.
• The lightweight nature of the models, particularly Deep-NSR, makes them suitable for wearable devices and telemedicine applications.

Conclusions:

• Deep-SR and Deep-NSR are effective and lightweight deep learning models for accurate cardiac arrhythmia recognition from ECG.
• These models offer potential for long-term cardiac monitoring via wearable devices and improved telemedicine diagnostics.
• The adoption of these models can reduce healthcare costs and physician workload.