Multi-Class Arrhythmia Detection from PPG Signals Based on VGG-BiLSTM Hybrid Deep Learning Model

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Arrhythmia poses a significant cardiovascular risk.
• Photoplethysmography (PPG) offers a noninvasive method for cardiac monitoring.
• Current PPG methods often struggle with multi-class arrhythmia classification.

Purpose of the Study:

• To develop a deep learning model for classifying six types of cardiac arrhythmias using PPG signals.
• To enhance PPG signal analysis by incorporating signal derivatives.
• To improve the accuracy of noninvasive arrhythmia detection.

Main Methods:

• A hybrid VGG-BiLSTM deep learning architecture was employed.
• Raw PPG signals were augmented with first and second derivatives.
• Stratified data splitting was used to manage class imbalance.
• A publicly available dataset of 46,827 PPG segments was utilized.

Main Results:

• The VGG-BiLSTM model achieved an overall accuracy of 88.7%.
• Sensitivity, specificity, and F1 score were reported as 78.5%, 97.6%, and 80.5%, respectively.
• The method demonstrated effectiveness in categorizing various arrhythmia types.

Conclusions:

• The proposed deep learning approach effectively classifies multiple arrhythmia types from PPG data.
• The VGG-BiLSTM model shows promise for advanced noninvasive cardiac rhythm monitoring.
• This technique offers a potential improvement over existing binary classification PPG methods.