Enhancing cardiovascular disease classification in ECG spectrograms by using multi-branch CNN

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Cardiovascular diseases (CVDs) pose a significant global health threat, necessitating accurate early detection.
• Electrocardiograms (ECGs) are crucial for diagnosing CVDs, but manual interpretation can be challenging.
• Artificial intelligence (AI), particularly deep learning (DL), shows promise for automated CVD classification from ECG data.

Purpose of the Study:

• To propose and compare the performance of 1D CNN, 2D CNN, and multi-branch CNN (MB-CNN) models for classifying CVDs from ECGs.
• To evaluate the effectiveness of converting 1D ECGs into 2D spectrograms for improved classification accuracy.
• To develop a highly accurate AI model for the early detection of various cardiovascular conditions.

Main Methods:

• One-dimensional (1D) convolutional neural networks (CNNs) were applied to raw ECG signals.
• 1D ECG signals were transformed into 2D-ECG spectrograms using continuous wavelet transform (CWT).
• Two-dimensional (2D) CNN and a proposed multi-branch CNN (MB-CNN) were utilized to classify these spectrograms.

Main Results:

• The 1D CNN achieved a maximum performance of 97.60%.
• The 2D CNN, applied to CWT-generated spectrograms, reached a maximum accuracy of 98.46%.
• The proposed MB-CNN model demonstrated superior performance, achieving an average test accuracy of 99.34% for classifying five CVD types and 99.22% for 5-class ECG classification.

Conclusions:

• Deep learning models, especially MB-CNN, significantly enhance the accuracy of CVD classification from ECG data.
• Converting 1D ECGs to 2D spectrograms and employing advanced CNN architectures improves diagnostic performance.
• The developed MB-CNN model shows high potential for accurate and early prediction of cardiovascular diseases, aiding in mortality reduction.