Short Single-Lead ECG Signal Delineation-Based Deep Learning: Implementation in Automatic Atrial Fibrillation...

Area of Science:

• Cardiology
• Artificial Intelligence
• Signal Processing

Background:

• Manual interpretation of electrocardiogram (ECG) signals is time-consuming and subjective.
• Current deep learning (DL) methods for ECG analysis are often limited to specific abnormalities.
• Generalizing automatic ECG interpretation requires accurate waveform delineation.

Purpose of the Study:

• To develop a generalized automatic interpretation model for ECG signals.
• To delineate ECG waveforms (P-wave, QRS-complex, T-wave) for improved analysis.
• To identify atrial fibrillation (AF) using delineated ECG morphology and medical knowledge rules.

Main Methods:

• A deep learning model was developed to delineate ECG waveforms.
• The delineation model was trained on the QT database and validated using The Lobachevsky University Database.
• Atrial fibrillation (AF) identification was performed by combining waveform delineation with medical rules, focusing on RR irregularities and P-wave absence.

Main Results:

• The ECG delineation model demonstrated high performance: 98.91% sensitivity, 99.01% precision, 99.79% specificity, 99.79% accuracy, and 98.96% F1 score.
• Testing on 4058 normal sinus rhythm and 1804 AF records from three datasets confirmed the model's ability to identify AF.
• The approach yielded high negative and positive predictive values for AF detection.

Conclusions:

• The proposed model effectively delineates ECG waveforms, enabling accurate automatic interpretation.
• Combining ECG delineation with medical rules provides a robust method for atrial fibrillation (AF) identification.
• This approach offers a significant contribution to improving AF diagnosis through automated ECG analysis.