Two-Stage Dynamic Synergistic Segmentation Method for Myocardial Pathology

Area of Science:

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

Background:

• Accurate segmentation of myocardial scar and edema from multi-sequence cardiac magnetic resonance (MS-CMR) is crucial for myocardial infarction assessment.
• Challenges include heterogeneous imaging characteristics, class imbalance, and ambiguous pathological regions.

Purpose of the Study:

• To propose a dynamic synergistic segmentation network (DSS-Net) for robust myocardial pathology segmentation in MS-CMR.
• To address limitations in current segmentation methods for scar and edema detection.

Main Methods:

• A coarse-to-fine strategy with initial myocardium segmentation for anatomical priors.
• A Modality Dynamic Fusion Module (MDFM) for adaptive emphasis on relevant imaging data.
• A Stage Feature Aggregation Module (SFAM) to improve cross-stage feature interaction and lesion representation.

Main Results:

• DSS-Net achieved competitive performance on the MyoPS 2020 and MyoPS 2024 datasets.
• Dice scores of 0.706 for scar and 0.753 for edema were obtained on the MyoPS 2020 dataset.
• The method demonstrated a balanced trade-off between sensitivity and specificity compared to state-of-the-art approaches.

Conclusions:

• Combining anatomical guidance with pathology-aware multi-modal learning is effective for myocardial pathology segmentation.
• DSS-Net offers a promising strategy for accurate and robust assessment of myocardial infarction from MS-CMR images.