Dual Coordinate Attention (DCA) Network for Accurate Cerebral Vascular Endothelium Segmentation in OCT Images

Accurate cerebral vascular endothelium segmentation in Optical Coherence Tomography (OCT) images is crucial for cerebrovascular disease assessment, yet faces challenges including laborious manual annotation, high-quality dataset needs, and limitations in existing attention mechanisms for unified feature modeling. This paper proposes a novel segmentation framework with a Dual Coordinate Attention (DCA) mechanism, validated on a self-constructed, meticulously annotated cerebrovascular OCT dataset. DCA facilitates robust feature interaction between Cartesian and polar coordinate representations, effectively capturing complementary structural cues from both domains to enhance endothelial features and suppress noise. Extensive experiments demonstrate the framework's superior performance (Dice and HD95) over traditional baseline models. Ablation studies confirm the DCA module's benefits and pinpoint optimal deployment. Leveraging dedicated data curation and novel DCA, this work provides an an accurate, robust automated segmentation tool for cerebral vascular endothelium in OCT images, promising to aid cerebrovascular condition assessment and monitoring.