Multiclass Arrhythmia Detection and Classification From Photoplethysmography Signals Using a Deep Convolutional...

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Photoplethysmography (PPG) signals are increasingly used for detecting atrial fibrillation.
• Classifying multiple types of arrhythmias using PPG signals is less explored.
• This study addresses the gap by investigating PPG for multiclass arrhythmia classification.

Purpose of the Study:

• To explore the feasibility of using PPG signals with a deep convolutional neural network (DCNN) for multiclass arrhythmia classification.
• To develop and validate a DCNN model for identifying various heart rhythm disorders from PPG data.

Main Methods:

• Simultaneously collected ECG and PPG signals from patients undergoing radiofrequency ablation for arrhythmias.
• Developed a DCNN to classify 6 types of rhythms (sinus rhythm, PVC, PAC, VT, SVT, AF) using 10-second PPG waveforms.
• Evaluated performance using accuracy, sensitivity, specificity, PPV, NPV, and AUC against cardiologist-annotated ECGs.

Main Results:

• The DCNN achieved 85.0% overall accuracy in classifying 6 arrhythmia types on an independent test set.
• Microaverage AUC reached 0.978, indicating high classification performance.
• Average sensitivity, specificity, PPV, and NPV were 75.8%, 96.9%, 75.2%, and 97.0%, respectively.

Conclusions:

• Classifying multiclass arrhythmias from PPG signals using deep learning is feasible.
• This non-invasive approach shows potential for population-based arrhythmia screening.
• The method may offer promise for long-term arrhythmia surveillance and management.