Transformers-Based Neural Network for Cardiac Infarction Segmentation in Delayed-Enhancement MRI
Résumé
Accurately and robustly segmenting myocardial infarction (MI) is crucial for clinical diagnosis of cardiac diseases, treatment and planning. In this study, we propose a novel deep learning model specifically designed for automatic segmentation of MI in Late Gadolinium Enhancement cardiac MRI (LGE-MRI). LGE-MRI is widely used in clinical practice to quantify MI and plays a vital role in treatment decisions. However, due to the presence of high anisotropy and inhomogeneities in LGE-MRI, accurately segmenting the infarcted tissue poses significant challenges.
Our approach introduces the use of U-Net transformers for MI segmentation. By leveraging the power of transformerbased architectures, our model achieves competitive results. We evaluated our method on the 2020 MICCAI EMIDEC challenge dataset and obtained a dice score of 91.33% for myocardium segmentation and 74.41% for infarction segmentation. These results demonstrate the effectiveness of our approach, showcasing its superiority over existing state-of-the-art methods in the EMIDEC challenge.
Origine | Fichiers produits par l'(les) auteur(s) |
---|