Inter-Patient Congestive Heart Failure Detection Using ECG-Convolution-Vision Transformer Network.

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Congestive heart failure (CHF) presents life-threatening symptoms like dyspnea and fatigue.
• Electrocardiograms (ECGs) are crucial for CHF diagnosis but are prone to misinterpretation due to waveform complexity.
• Current machine learning approaches for automated CHF detection often lack robustness in noisy conditions and inter-patient variability.

Purpose of the Study:

• To develop and evaluate a novel deep learning model for accurate and robust detection of CHF from ECG signals.
• To address limitations in existing automated CHF detection methods, specifically their performance under noisy conditions and inter-patient generalization.

Main Methods:

• Proposed a hybrid deep learning architecture, the ECG-Convolution-Vision Transformer Network (ECVT-Net).
• ECVT-Net integrates Convolutional Neural Network (CNN) and Vision Transformer components for automatic feature extraction from ECGs.
• Evaluated model performance using an inter-patient experimental scheme and assessed robustness by introducing varying levels of noise to ECG data.

Main Results:

• Achieved a high accuracy of 98.88% in identifying CHF using the inter-patient scheme.
• Demonstrated significant noise robustness, maintaining effective performance even with the addition of noise to ECG signals.
• The ECVT-Net model effectively extracts high-dimensional abstract features from ECGs with minimal pre-processing.

Conclusions:

• ECVT-Net provides an effective and accurate method for identifying CHF from ECGs.
• The model's robustness to noise makes it suitable for real-world clinical applications where signal quality can vary.
• This approach advances automated CHF detection, offering improved reliability over existing methods.