Multi-Scale Attention Network for Myocardial Infarction Transmurality Classification in Late Gadolinium Enhancement...

The transmural extent of hyperenhancement on late gadolinium enhancement (LGE) cardiac magnetic resonance (CMR) imaging is a key marker of myocardial infarction severity and prognosis. Current visual assessment methods suffer from inter-observer variability and reliance on manual segmentation. In this paper, we propose a Multi-Scale Attention Network for Transmurality Classification (MSAN-TC) using LGE CMR images. MSAN-TC integrates convolutional neural networks (CNNs) and Transformer models with feature pyramid networks (FPN) and channel attention (CA) mechanisms, enabling accurate classification of infarction extent from weakly labeled data. Evaluated on 1,821 images from 315 patients, MSAN-TC achieves an overall accuracy of 86% in transmurality classification and an area under the curve (AUC) of 0.90 in detecting ≥50% transmural infarction, demonstrating high sensitivity (91%) and specificity (89%). This work represents a step towards automated, efficient, and clinically practical myocardial infarction assessment, providing a scalable solution for real-world applications.