.png){kind=icon}
.png){kind=icon}
Return
Learning contextual representations with copula function for medical image segmentation
.png){kind=icon}
PRE
AI
DOI：10.1016/j.bspc.2023.104900.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Long-range contextual information modeling plays an important role in medical image segmentation. Existing approaches often ignore local continuous information or the attention missing. In this paper, we propose a novel approach named CouplaNet for efficiently modeling dense 3D contextual dependencies. CouplaNet consists of two core modules, namely Relationship Modelling (RM) and Contextual Aggregation (CA). RM leverages the GCN to model the relationships of elements and yields the long-range contextual representation for each dimension individually. Then CA is introduced to perform the aggregation for the ensemble of the representation of each dimension according to the coupla property in the Sklar's theorem. We conducted extensive experiments across six medical image segmentation tasks, including nodule segmentation, skin lesion segmentation, liver tumor segmentation, optic disc and cup segmentation, thyroid nodule segmentation, and automated cardiac diagnosis. Through these experiments, we were able to thoroughly evaluate our contributions and demonstrate the effectiveness of our approach.
Keywords:
Medical image segmentation
3D contextual information
Long-range dependencies
Journal
IF:
4.9
Papers:
9.7K
Citations:
2.4W