An Arrhythmia classification approach via deep learning using single-lead ECG without QRS wave detection.

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Arrhythmia is a common, life-threatening cardiac disorder.
• Traditional arrhythmia detection relies on QRS wave detection.
• Wearable devices offer continuous ECG monitoring for prompt arrhythmia detection.

Purpose of the Study:

• To classify distinct arrhythmias using a convolutional neural network (CNN) without relying on QRS wave detection.
• To evaluate the impact of ECG signal duration (5-s vs. 10-s segments) on arrhythmia classification accuracy.
• To assess the feasibility of using short ECG recordings for arrhythmia detection with wearable devices.

Main Methods:

• Utilized a one-dimensional CNN model for arrhythmia classification.
• Trained and compared CNN models using 5-second and 10-second ECG segments from PhysioNet databases.
• Classified Normal Sinus Rhythm (NSR) and various arrhythmias including Atrial Fibrillation (AFIB), Atrial Flutter (AFL), Wolff-Parkinson-White syndrome (WPW), Ventricular Fibrillation (VF), Ventricular Tachycardia (VT), Ventricular Flutter (VFL), Mobitz II AV Block (MII), and Sinus Bradycardia (SB).

Main Results:

• The CNN model successfully differentiated between NSR and multiple types of arrhythmias.
• Both 5-s and 10-s ECG segments yielded comparable classification accuracy.
• Achieved an average classification accuracy of 97.31% for arrhythmia detection.

Conclusions:

• CNNs can effectively classify arrhythmias without QRS wave detection.
• Short ECG segments (5-s) are feasible for arrhythmia detection, aligning with wearable device practicality.
• The method supports rapid arrhythmia detection for clinical utility in emergency situations.