Cardiac Arrhythmia Detection Leveraging Deep Learning Convolutional Neural Network.

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Signal Processing

Background:

• Cardiac arrhythmias, including atrial fibrillation (AF), are increasingly common and elevate stroke and heart failure risks.
• Photoplethysmography (PPG) signals offer a non-invasive method for monitoring cardiovascular health.

Purpose of the Study:

• To develop and evaluate a deep convolutional neural network (DCNN) for differentiating atrial fibrillation (AF) from sinus rhythm (SR) and premature contractions using PPG signals.

Main Methods:

• A 1D VGG16-Net model was trained on 46,827 10-second raw PPG signal segments from 91 patients.
• The model was evaluated using 10-fold cross-validation and independent testing datasets.

Main Results:

• The DCNN achieved a balanced accuracy of 0.895 ± 0.03 during cross-validation.
• Independent testing yielded a balanced accuracy of 0.8279 in differentiating between SR, premature contractions, and AF.

Conclusions:

• The DCNN demonstrates capability in distinguishing between normal sinus rhythm, premature contractions, and atrial fibrillation using PPG data.
• Further improvements in model accuracy are needed, but PPG-based screening for AF and premature contractions is validated.