Premature Ventricular Contraction Detection from Ambulatory ECG Using Recurrent Neural Networks

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Premature ventricular contractions (PVCs) are typically benign but frequent occurrences can lead to arrhythmia-induced cardiomyopathy, increasing risks of heart failure and mortality.
• High PVC counts are recognized as predictors for severe arrhythmias.
• Wearable devices offer convenient, continuous monitoring for PVC detection in daily life.

Purpose of the Study:

• To develop a reliable and low-cost data analysis program for real-time PVC detection using data from wearable devices.
• To leverage deep learning for accurate identification of frequent premature beats.

Main Methods:

• Utilized recurrent neural networks (RNNs) with long short-term memory (LSTM) architecture for PVC detection.
• Validated the developed method using the MIT-BIH arrhythmia database.

Main Results:

• Achieved high detection accuracy rates ranging from 96% to 99% for PVCs.
• Demonstrated the efficacy of LSTM-based deep learning for analyzing large datasets from wearable monitors.

Conclusions:

• Deep learning models, specifically LSTM networks, provide an accurate and efficient method for real-time PVC detection.
• This approach facilitates risk stratification for severe arrhythmias and supports proactive cardiac health management using wearable technology.