University of Pittsburgh (Riek et al.)
Model weights not public. Contact creators for more information.
Clinically-informed modification of the ResNet-18 architecture for identifying occlusion myocardial infarction (OMI) -- a severe, often ST-elevation-negative heart attack caused by complete blockage of a coronary artery -- from a single 12-lead ECG. The network first learns lead-specific temporal features via 1xk temporal convolutions, then learns cross-lead spatial concordance/discordance (e.g. reciprocal ST changes) via a 12x1 spatial convolution placed after the residual blocks, with saliency maps highlighting the most relevant leads and waveform regions for explainability. Benchmarked against ResNet-18 and other CNN/random-forest baselines on a multisite real-world clinical dataset of 10,893 ECGs (OMI rate 6.5%), reaching a test AUROC of 0.889 and an average precision of 0.587, outperforming the compared models.
Architecture
Hybrid
Clinically-informed ResNet-18 variant: lead-specific temporal 1D convolutions (1xk kernels) through the residual blocks, followed by a cross-lead spatial convolution (12x1 kernel) to capture inter-lead concordance/discordance, with saliency-map explainability
Framework
PyTorch
Added to catalog
2026-08-12
Multisite Prehospital/Emergency-Department 12-lead ECG OMI Cohort (Pittsburgh)
Multisite 12-lead ECGs from patients with suspected acute coronary syndrome presenting via emergency medical services or the emergency department; 6.5% of ECGs are labeled occlusion myocardial infarction (OMI) by expert angiographic adjudication.
Binary classification of occlusion myocardial infarction (OMI) vs. non-OMI/control from a single 12-lead ECG