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

Area of Science:

• Cardiovascular Imaging
• Artificial Intelligence in Medicine
• Medical Image Analysis

Background:

• Late gadolinium enhancement (LGE) cardiac magnetic resonance (CMR) imaging is crucial for assessing myocardial infarction (MI) severity and prognosis.
• Current visual assessment of LGE images has limitations, including inter-observer variability and reliance on manual segmentation.

Purpose of the Study:

• To develop and evaluate an automated method for classifying the transmural extent of myocardial infarction using LGE CMR images.
• To address the limitations of manual assessment by introducing a novel deep learning model.

Main Methods:

• A Multi-Scale Attention Network for Transmurality Classification (MSAN-TC) was proposed, integrating CNNs, Transformer models, feature pyramid networks (FPN), and channel attention (CA).
• The model was trained and evaluated on 1,821 LGE CMR images from 315 patients, utilizing weakly labeled data for classification.

Main Results:

• MSAN-TC achieved an overall accuracy of 86% in transmurality classification.
• The model demonstrated a high area under the curve (AUC) of 0.90 for detecting ≥50% transmural infarction.
• High sensitivity (91%) and specificity (89%) were observed in the classification task.

Conclusions:

• The MSAN-TC model offers an automated, efficient, and clinically practical solution for assessing myocardial infarction extent.
• This deep learning approach represents a significant step towards scalable and reliable MI assessment in real-world clinical settings.