Classification of QRS complexes to detect Premature Ventricular Contraction using machine learning techniques.

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning

Background:

• Premature Ventricular Contractions (PVCs) are common cardiac arrhythmias requiring accurate detection for patient care.
• Manual electrocardiogram (ECG) analysis for PVCs is time-consuming and can be challenging, impacting healthcare efficiency.
• Automated detection of PVCs can significantly aid in managing cardiac health and reducing expert workload.

Purpose of the Study:

• To develop and evaluate a machine learning approach for automated PVC detection using ECG data.
• To assess the performance of various classifiers, including deep learning models, without feature extraction or cross-validation.
• To identify the most effective model for accurate and efficient PVC identification.

Main Methods:

• Utilized the MIT-BIH Arrhythmia database for training and testing machine learning models.
• Implemented and compared six classifiers: Decision Tree, Random Forest, LSTM, Bidirectional LSTM, ResNet-18, MobileNetv2, and ShuffleNet.
• Conducted experiments on both original and balanced datasets to evaluate model robustness.

Main Results:

• MobileNetv2 demonstrated superior performance in PVC detection across both experimental datasets.
• Achieved high accuracy rates of 99.90% on the original dataset and 99.00% on the balanced dataset.
• The study highlights the effectiveness of deep learning models in classifying PVCs without traditional feature engineering.

Conclusions:

• The proposed machine learning approach, particularly MobileNetv2, offers a highly accurate and efficient method for PVC detection.
• This automated system can reduce the burden on healthcare professionals analyzing ECGs.
• Further research is needed to understand the explainability of deep learning model decisions in cardiac diagnostics.