ECG-SMART-NET: A Deep Learning Architecture for Precise ECG Diagnosis of Occlusion Myocardial Infarction

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Occlusion myocardial infarction (OMI) is a critical heart attack subtype requiring urgent intervention.
• Visual ECG interpretation misses two-thirds of OMI cases, delaying life-saving treatment.
• Current AI approaches, including random forests and CNNs, show promise for OMI detection.

Purpose of the Study:

• Develop and evaluate ECG-SMART-NET, a novel AI model for OMI identification using 12-lead ECGs.
• Improve the accuracy and speed of OMI diagnosis, particularly in challenging cases.
• Enhance the clinical utility of ECGs for detecting critical coronary artery blockages.

Main Methods:

• A modified ResNet-18 architecture, ECG-SMART-NET, was developed.
• The model incorporates temporal convolutional layers (1xk kernels) for lead-specific feature learning.
• Spatial convolutional layers (12x1 kernels) capture cross-lead relationships after residual blocks.

Main Results:

• ECG-SMART-NET demonstrated superior performance on a large, multi-site clinical dataset (10,393 ECGs).
• The model achieved a high test Area Under the Curve (AUC) of 0.953 [0.921, 0.978] for OMI classification.
• Performance surpassed the original ResNet-18 and other state-of-the-art models.

Conclusions:

• ECG-SMART-NET significantly outperforms state-of-the-art random forest models for OMI prediction.
• The proposed architecture is better suited for ECG analysis than the standard ResNet-18.
• ECG-SMART-NET offers a promising advancement in AI-driven OMI detection.