فهرست نشریات علمی دسترسی آزاد ایران

 Background and Objectives: Early revascularization of occluded coronary arteries in patients with ST-elevation myocardial infarction (STEMI) has been shown to reduce mortality. Currently, physicians rely on electrocardiographic (ECG) features to identify the most likely location of occlusion in coronary arteries. We sought to more accurately predict these culprit arteries using deep learning.   Methods: This retrospective case-control study was conducted to diagnose culprit lesion based on ECG using artificial intelligence (AI) in patients with STEMI referred to Imam Khomeini Hospital and Tehran Heart Center, Tehran, Iran. In this study, 804 patients who underwent coronary angiography (CAG) in STEMI setting were included. Subsequently, the ECG and angiography data of these patients were reevaluated, and AI experts analyzed the ECG and CAG using the EfficientNet_B7 algorithm, a deep learning approach that employs convolutional neural networks for high accuracy and computational efficiency. Finally, the presented model was analyzed.   Results: This study evaluated a cohort of 804 patients. The proposed architecture, based on the EfficientNet_B7 algorithm, effectively integrates Transfer Learning and EfficientNet techniques to optimize data pre-processing prior to network input. A learning rate of 1×10−4 was employed to facilitate gradual weight updates, while a batch size of 32 was selected, thereby minimizing the risks of overfitting and underfitting. The training dataset was divided into 70% for training, 15% for validation, and 15% for testing, with the network's error function determined using Binary Cross Entropy loss. These findings underscore the potential of AI-driven approaches in enhancing diagnostic accuracy. Utilizing an AI model for ECG signal analysis, we achieved an impressive overall prediction accuracy of 98%. The model demonstrated a sensitivity of 96% and precision of 98% in identifying the left anterior descending (LAD) artery as the culprit vessel.   Conclusion: Deep learning has the potential to enhance the diagnosis of culprit vessel in patients with STEMI. Our model successfully detected LAD occlusion in STEMI with 98% accuracy, 96% sensitivity, and 98% F1-score.