BlendNet: a blending-based convolutional neural network for effective deep learning of electrocardiogram signals

Area of Science:

• Cardiology
• Artificial Intelligence
• Signal Processing

Background:

• Deep learning (DL) models, including Convolutional Neural Networks (CNNs), are increasingly used for diagnosing cardiovascular diseases from ElectroCardioGram (ECG) signals.
• Traditional methods often convert 1-D ECG signals into 2-D scalogram images using Continuous Wavelet Transform (CWT) for DL analysis, but this can limit feature extraction.

Purpose of the Study:

• To introduce BlendNet, a novel DL architecture designed to enhance feature extraction from ECG signals.
• To improve the accuracy of cardiovascular disease detection by utilizing an innovative alpha blending technique.

Main Methods:

• The proposed BlendNet architecture employs alpha blending to combine features from both a CWT-generated scalogram image and its binary version.
• ECG signals are transformed into scalogram images, and both original and binary versions are processed through convolutional and pooling layers.
• The extracted features are then blended using a flexible parameter (α) before classification by a dense layer.

Main Results:

• BlendNet, particularly with α = 0.7, demonstrated superior performance in ECG classification compared to traditional and state-of-the-art methods.
• Experiments utilized 162 ECG recordings from the PhysioNet database.

Conclusions:

• Alpha blending in BlendNet generates richer composite feature sets, leading to improved classification accuracy for cardiovascular diseases.
• The BlendNet architecture offers flexibility in dense layer settings and can be integrated with machine learning algorithms for faster convergence.