Digitization and Classification of Electrocardiograms for Automated Cardiovascular Disease Diagnosis Using Machine Learning and Deep Learning Models

Le résumé fourni par la source

Electrocardiograms (ECG) enable the straightforward identification of cardiovascular diseases (CVD). However, the complexity of ECG graphs often challenges physicians in making precise and confident diagnoses. This study aims to digitize electrocardiograms using a publicly available ECG image dataset from the Mendeley database and classify the signals into four categories: normal, arrhythmia, myocardial infarction (MI), and past myocardial infarction (PMI). The proposed framework includes a comprehensive preprocessing pipeline involving greyscale conversion, noise elimination, segmentation, and normalization. The classification process employs advanced deep learning models like EfficientNet-1D and ResNet-1D alongside machine learning (ML) models such as XGBoost, Support Vector Machines, Logistic Regression, and k-Nearest Neighbours (kNN). The models were evaluated using performance metrics, including F1-score, accuracy, recall, and precision. The Voting Ensemble Classifier outperformed achieving the highest overall accuracy of 93.2%. Among the deep learning models, ResNet-1D achieved an accuracy of 86.6%, while EfficientNet-1D attained 92.2%, demonstrating strong performance in ECG classification. ResNet-1D excelled in detecting normal and arrhythmic signals, while EfficientNet-1D demonstrated superior effectiveness in identifying myocardial infarctions. Both models exhibited high precision and recall, with ResNet-1D achieving exceptional success. This study underscores the importance of digitizing printed ECG records for advanced medical diagnostics and highlights the role of performance metrics in evaluating model efficacy. The proposed method enhances ECG classification accuracy and is well-suited for real-time clinical applications and automated cardiac disease detection systems.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.