Cervical Cell Segmentation Method Based on Global Dependency and Local Attention

The refined segmentation of nuclei and the cytoplasm is most challenging task in automation cervical cell screening. U-Shape network structure has demonstrated great superiority field biomedical imaging. However, classical U-Net cannot effectively utilize mixed domain information contextual information, fails to achieve satisfactory results this task. To address above problems, a module based on global dependency local attention (GDLA) for modeling features refinement, proposed study. It consists three components computed parallel, which are module, spatial channel module. models capture priori knowledge cells, such as positional dependence cytoplasm, closure uniqueness nuclei. combines extract boundary refine target boundaries. modules used provide adaption input make it easy identify subtle but dominant differences similar objects. Comparative ablation experiments conducted Herlev dataset, experimental demonstrate effectiveness method, surpasses popular existing attention, hybrid context networks terms metrics, achieving better performance than previous advanced methods.