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

Nathan T. Riek, Murat Akçakaya+11 more|arXiv (Cornell University)|May 8, 2024

ECG Monitoring and AnalysisMedicine3 citations

TL;DR

ECG-SMART-NET is a deep learning architecture that enhances occlusion myocardial infarction (OMI) detection in 12-lead ECGs by integrating temporal and spatial feature learning through a modified ResNet-18 with 1×k and 12×1 kernel convolutions. It achieved a test AUC of 0.953 on a real-world multisite dataset of 10,393 ECGs, outperforming both ResNet-18 and state-of-the-art random forest models.

ABSTRACT

Objective: In this paper we develop and evaluate ECG-SMART-NET for occlusion myocardial infarction (OMI) identification. OMI is a severe form of heart attack characterized by complete blockage of one or more coronary arteries requiring immediate referral for cardiac catheterization to restore blood flow to the heart. Two thirds of OMI cases are difficult to visually identify from a 12-lead electrocardiogram (ECG) and can be potentially fatal if not identified quickly. Previous works on this topic are scarce, and current state-of-the-art evidence suggests both feature-based random forests and convolutional neural networks (CNNs) are promising approaches to improve ECG detection of OMI. Methods: While the ResNet architecture has been adapted for use with ECG recordings, it is not ideally suited to capture informative temporal features within each lead and the spatial concordance or discordance across leads. We propose a clinically informed modification of the ResNet-18 architecture. The model first learns temporal features through temporal convolutional layers with 1xk kernels followed by a spatial convolutional layer, after the residual blocks, with 12x1 kernels to learn spatial features. Results: ECG-SMART-NET was benchmarked against the original ResNet-18 and other state-of-the-art models on a multisite real-word clinical dataset that consists of 10,393 ECGs from 7,397 unique patients (rate of OMI =7.2%). ECG-SMART-NET outperformed other models in the classification of OMI with a test AUC of 0.953 [0.921, 0.978]. Conclusion and Significance: ECG-SMART-NET can outperform the state-of-the-art random forest for OMI prediction and is better suited for this task than the original ResNet-18 architecture.

Motivation & Objective

• To address the clinical challenge of detecting occlusion myocardial infarction (OMI) in 12-lead ECGs, which is missed in two-thirds of cases due to subtle or non-specific patterns.
• To develop a deep learning model that captures both temporal dynamics within individual ECG leads and spatial relationships across leads more effectively than standard architectures.
• To outperform existing state-of-the-art models, including ResNet-18 and random forest classifiers, in OMI diagnosis using real-world clinical ECG data.
• To create a clinically informed neural network architecture that improves early detection of OMI, a life-threatening condition requiring urgent reperfusion therapy.

Proposed method

• The model is a modified ResNet-18 architecture with temporal convolutional layers using 1×k kernels applied before residual blocks to extract temporal features from each ECG lead.
• After the residual blocks, a spatial convolutional layer with 12×1 kernels is applied to learn inter-lead spatial concordance or discordance across the 12 leads.
• The architecture preserves the residual learning framework of ResNet-18 while introducing specialized convolutions tailored to ECG signal structure.
• The model is trained end-to-end on a multisite clinical dataset of 10,393 ECGs from 7,397 unique patients with a 7.2% OMI prevalence rate.
• Performance is evaluated using area under the ROC curve (AUC), with comparisons to baseline models including original ResNet-18 and a state-of-the-art random forest classifier.

Experimental results

Research questions

• RQ1 Can a deep learning model that explicitly models temporal and spatial features in ECGs improve OMI detection accuracy compared to standard architectures?
• RQ2 Does a clinically informed modification of ResNet-18 with specialized 1×k and 12×1 convolutions outperform the original ResNet-18 in OMI classification?
• RQ3 How does ECG-SMART-NET compare to state-of-the-art models such as random forests and standard CNNs in detecting OMI on real-world clinical ECG data?
• RQ4 To what extent do temporal and spatial feature learning components contribute independently and jointly to improved diagnostic performance?

Key findings

• ECG-SMART-NET achieved a test AUC of 0.953 (95% CI: 0.921–0.978) on a multisite clinical dataset of 10,393 ECGs, demonstrating superior performance in OMI detection.
• The model significantly outperformed the original ResNet-18 architecture, which served as a baseline, in terms of AUC and diagnostic accuracy.
• ECG-SMART-NET surpassed the performance of a state-of-the-art random forest model, which was previously considered a strong benchmark for OMI detection.
• The integration of 1×k temporal convolutions followed by 12×1 spatial convolutions enabled the model to effectively capture both lead-specific dynamics and inter-lead patterns critical for OMI diagnosis.