Premature beats detection based on a novel convolutional neural network.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Cardiology

Background:

• Automatic detection of premature beats in electrocardiogram (ECG) recordings is crucial for clinical diagnosis.
• Premature beats, including premature ventricular contractions (PVCs) and supraventricular premature beats (SPBs), require accurate identification for effective patient management.

Purpose of the Study:

• To propose a novel deep learning model, ECGDet, for the automatic detection of PVCs and SPBs in single-lead long-term ECGs.
• To evaluate the performance of ECGDet on established arrhythmia databases and a competitive challenge.

Main Methods:

• The ECGDet model utilizes a convolutional neural network and squeeze-and-excitation network architecture.
• A novel loss calculation method was implemented during model training.
• The model was trained and validated using five-fold cross-validation on the MIT-BIH arrhythmia database (MITDB) and further tested on the CPSCDB.

Main Results:

• ECGDet achieved an average F1 score of 92.6% for PVC detection and 72.2% for SPB detection on the MITDB.
• The model secured 2nd place in PVC detection and 7th place in SPB detection at the China Physiological Signal Challenge (2020).

Conclusions:

• The proposed ECGDet model demonstrates effective automatic detection of premature heartbeats from long-term ECG signals.
• This deep learning approach eliminates the need for manual feature extraction, offering significant potential for clinical applications and long-term ECG analysis.