A novel proposed CNN-SVM architecture for ECG scalograms classification.

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence in Healthcare

Background:

• Increasing sudden cardiac deaths necessitate advanced diagnostic tools.
• Automatic electrocardiogram (ECG) classification is vital for timely diagnosis and treatment.
• Deep learning offers automated feature extraction for complex signal analysis.

Purpose of the Study:

• To propose a novel 34-layer convolutional neural network (CNN) architecture for ECG type detection.
• To evaluate the CNN's ability to automatically extract features from ECG images.
• To enhance classification accuracy by combining the CNN with Support Vector Machines (SVM).

Main Methods:

• One-dimensional ECG signals were transformed into images (scalograms) using Continuous Wavelet Transform (CWT).
• A novel 34-layer CNN was developed and compared against AlexNet and SqueezeNet.
• The CNN was utilized as a deep feature extractor and combined with SVM for final classification, employing cross-validation.

Main Results:

• The proposed CNN architecture demonstrated superior performance compared to AlexNet and SqueezeNet in classifying ECG images.
• Continuous Wavelet Transform (CWT) and cross-validation were identified as optimal pre-processing and data splitting methods.
• The combined CNN-SVM model achieved a highest classification accuracy of 99.21% using CWT.

Conclusions:

• The proposed CNN-SVM framework effectively classifies ECG types with high accuracy.
• This automated system shows significant potential as a clinical aid for ECG interpretation.
• The study highlights the efficacy of deep learning, specifically CNNs and SVMs, in advancing cardiac diagnostics.