Enhancing ECG disease detection accuracy through deep learning models and P-QRS-T waveform features.

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Cardiovascular diseases (CVDs) are the leading cause of global mortality.
• Electrocardiograms (ECGs) are crucial for non-invasive heart condition diagnosis.
• Existing ECG research often focuses on binary or arrhythmia classification, necessitating advanced multi-class models.

Purpose of the Study:

• To develop a robust deep learning method for multi-class classification of various heart abnormalities using ECG data.
• To enhance the accuracy of diagnosing specific conditions like Hypertrophy, Conduction Disturbance, Myocardial Infarction, and ST-T Changes.

Main Methods:

• Utilized the PTB-XL ECG database, applying Butterworth bandpass and Discrete Wavelet Transform (DWT) db-8 filtering.
• Extracted morphological features (P-QRS-T intervals and amplitudes) from R-peak detected signals.
• Employed Synthetic Minority Oversampling Technique for Nominal and Continuous (SMOTE-NC) for data balancing, followed by Convolutional Neural Network (CNN) and Deep Neural Network (DNN) models with 5-fold cross-validation.

Main Results:

• The Deep Neural Network (DNN) model achieved a mean accuracy of 84% (±0.01), outperforming the Convolutional Neural Network (CNN) model's 81% (±0.03) accuracy.
• Hypertrophy (HYP) classification demonstrated high consistency, reaching up to 98% accuracy.
• Evaluated performance using F1 score, recall, precision, and accuracy across normal and four abnormal cardiac classes.

Conclusions:

• Combining advanced signal processing (DWT) with deep learning (CNN, DNN) effectively enables precise multi-class heart disease classification from ECGs.
• The P-QRS-T morphological features are valuable for distinguishing various cardiac abnormalities.
• The developed models show promise for future real-time clinical applications in cardiovascular diagnostics.