Premature ventricular contraction detection combining deep neural networks and rules inference

Area of Science:

• Cardiology
• Medical Informatics
• Artificial Intelligence in Medicine

Background:

• Premature ventricular contractions (PVCs) are common arrhythmias originating from ectopic heartbeats.
• PVCs can escalate to severe cardiac conditions, necessitating accurate detection.
• Computer-aided detection of PVCs is crucial for clinical and outpatient electrocardiogram (ECG) settings.

Purpose of the Study:

• To develop and evaluate a novel approach for accurate PVC detection.
• To combine deep neural networks with rule inference for enhanced diagnostic performance.
• To assess the method's effectiveness and generalization using diverse ECG databases.

Main Methods:

• A hybrid approach integrating deep neural networks and rule-based inference was developed.
• The method was validated on the MIT-BIH arrhythmia database (MIT-BIH-AR).
• Generalization capability was tested on the extensive Chinese Cardiovascular Disease Database (CCDD).

Main Results:

• On the MIT-BIH-AR database, the method achieved 99.41% accuracy, with 97.59% sensitivity and 99.54% specificity.
• Performance on the CCDD dataset (over 140,000 recordings) demonstrated 98.03% accuracy, 96.42% sensitivity, and 98.06% specificity.
• The proposed method outperformed existing techniques in PVC detection.

Conclusions:

• The combined deep neural network and rule inference approach offers a highly effective solution for PVC detection.
• The method demonstrates strong generalization capabilities across different datasets.
• This technique holds significant potential for improving the diagnosis and management of cardiac arrhythmias.