MyoPS-Net

MyoPS-Net segments myocardial pathologies from multi-sequence cardiac magnetic resonance (CMR) images to support diagnosis and treatment planning of myocardial infarction.

Key Features:

• End-to-End Deep Neural Network: An end-to-end deep neural network that integrates five different CMR sequences for comprehensive myocardial pathology analysis.
• Flexible Architecture: A flexible architecture that extracts and fuses cross-modal features from varying numbers of CMR images and complex combinations of modalities.
• Pathology-Specific Output Branches: Dedicated output branches targeting different myocardial pathologies to enhance segmentation specificity.
• Anatomical Knowledge Integration: A module that regularizes myocardium consistency and localizes pathologies to improve segmentation quality.
• Inclusiveness Loss: An inclusiveness loss function that exploits relationships between myocardial scars and edema to inform segmentation.
• Performance Evaluation: Evaluated on a private dataset of 50 paired multi-sequence CMR images and the MICCAI2020 MyoPS Challenge public dataset, demonstrating state-of-the-art performance across scenarios.
• Robustness to Incomplete Data: Extensive experiments assessing performance with incomplete CMR sequence combinations demonstrate improved generalizability.

Scientific Applications:

• Myocardial pathology segmentation: Delineation of myocardial scars and edema from multi-sequence CMR to support diagnosis and treatment planning for myocardial infarction.
• Multi-sequence CMR research: Study of cross-modal feature fusion and algorithm performance under complex modality combinations and incomplete-data scenarios.

Methodology:

An end-to-end deep learning framework that integrates five CMR sequences, performs cross-modal feature extraction and fusion with a flexible architecture, uses pathology-specific output branches, incorporates an anatomical-knowledge module to regularize myocardium consistency and localize pathologies, and employs an inclusiveness loss to exploit relationships between myocardial scars and edema.