Coarse-Grained Mask Regularization for Microvascular Obstruction Identification from Non-contrast Cardiac Magnetic Resonance

Identification of microvascular obstruction (MVO) in acute myocardial infarction patients is critical for prognosis and has a direct link to mortality risk. Current approaches using late gadolinium enhancement (LGE) for contrast-enhanced cardiovascular magnetic resonance (CMR) pose risks to the kidney and may not be applicable to many patients. This highlights the need to explore alternative non-contrast imaging methods, such as cine CMR, for MVO identification. However, the scarcity of datasets and the challenges in annotation make the MVO identification in cine CMR challenging and remain largely under-explored. For this purpose, we propose a non-contrast MVO identification framework in cine CMR with a novel coarse-grained mask regularization strategy to effectively utilize information from LGE annotations in training. We train and validate our model on a dataset comprising 680 cases. Our model demonstrates superior performance over competing methods in cine CMR-based MVO identification, proving its feasibility and presenting a novel and patient-friendly approach to the field. The code is available at https://github.com/code-koukai/MVO-identification .