توسعه يك مدل چند برچسبي تفسيرپذير مبتني بر يادگيري عميق به‌منظور تشخيص آريتميهاي قلبي با استفاده از سيگنال الكتروكارديوگرام

Accurate detection an‎d classification of cardiac arrhythmias from electrocardiogram (ECG) signals are of great clinical importance, as they enable the early identification of cardiac rhythm abnormalities an‎d timely therapeutic intervention. However, manual interpretation of ECG signals is a time-consuming process an‎d may be influenced by inter-observer variability among clinicians. Therefore, the development of intelligent artificial intelligence (AI)-based systems that provide both high diagnostic accuracy an‎d model interpretability has become an essential requirement in modern healthcare. In this dissertation, a novel deep learning architecture is proposed for the automatic classification of cardiac arrhythmias by integrating Convolutional Neural Networks (CNNs), Bidirectional Long Short-Term Memory (BiLSTM) networks, an‎d Transformer layers. The proposed model simultaneously captures morphological, temporal, an‎d spatial characteristics of ECG signals. The model was trained an‎d eva‎luated using the PhysioNet/Computing in Cardiology Challenge 2020 benchmark dataset, which contains more than 43,000 ECG recordings representing 27 different cardiac arrhythmia classes. Experimental results demonstrate that the proposed model achieved an overall accuracy of 91.24%, a macro F1-score of 83.01%, an‎d an area under the receiver operating characteristic curve (AUC) exceeding 0.98 for major cardiac arrhythmias. Furthermore, to enhance transparency an‎d clinical reliability, the proposed model was interpreted using the SHapley Additive exPlanations (SHAP) framework, which identifies the contribution of individual ECG features to the modelʹs predictions in an interpretable manner. The findings indicate that the proposed framework not only provides high diagnostic performance but also demonstrates strong potential for integration into intelligent clinical decision-support systems. Consequently, it represents a promising step toward the practical deployment of explainable deep learning models for automated cardiovascular disease diagnosis.

5

161832