Automated real-time method for ventricular heartbeat classification.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Wearable technologies require efficient, continuous electrocardiogram (ECG) monitoring.
• Single-lead ECG systems face challenges with signal quality due to user activity and non-standard leads.
• Accurate ventricular heartbeat classification is crucial for diagnosing cardiac conditions.

Purpose of the Study:

• To develop a fully automatic, real-time ventricular heartbeat classifier utilizing a single ECG lead.
• To create a computationally efficient algorithm suitable for resource-constrained wearable devices.
• To enable knowledge transfer between different ECG databases without retraining.

Main Methods:

• An Echo State Network (ESN) was employed for ECG signal classification.
• The method adheres to Association for the Advancement of Medical Instrumentation (AAMI) standards using an inter-patient scheme.
• Fast, computationally efficient feature extraction and classification were implemented for real-time performance.

Main Results:

• The classifier demonstrated high performance on the MIT-BIH arrhythmia and INCART databases.
• Sensitivity and precision reached 95.3% and 88.8% for modified lead II, and 90.9% and 89.2% for V1 lead.
• The single-lead classifier performance is comparable to state-of-the-art multi-lead algorithms.

Conclusions:

• The developed automated, single-lead, real-time classifier offers improved accuracy for ventricular heartbeats.
• The methodology is suitable for wearable long-term monitoring, even with unconventional electrode placements.
• This approach enhances the feasibility of advanced cardiac monitoring in diverse wearable applications.