BI-GCN: Boundary-Aware Input-Dependent Graph Convolution Network for Biomedical Image Segmentation
This work addresses segmentation challenges in biomedical imaging, such as limited receptive fields in convolutions, by integrating graph-based methods with attention mechanisms, though it appears incremental as an improvement over existing graph convolution approaches.
The paper tackles biomedical image segmentation by proposing a boundary-aware, input-dependent graph convolution network that leverages both region and boundary information for global semantic reasoning. It demonstrates state-of-the-art performance on polyp segmentation in colonoscopy images and optic disc/cup segmentation in fundus images.
Segmentation is an essential operation of image processing. The convolution operation suffers from a limited receptive field, while global modelling is fundamental to segmentation tasks. In this paper, we apply graph convolution into the segmentation task and propose an improved \textit{Laplacian}. Different from existing methods, our \textit{Laplacian} is data-dependent, and we introduce two attention diagonal matrices to learn a better vertex relationship. In addition, it takes advantage of both region and boundary information when performing graph-based information propagation. Specifically, we model and reason about the boundary-aware region-wise correlations of different classes through learning graph representations, which is capable of manipulating long range semantic reasoning across various regions with the spatial enhancement along the object's boundary. Our model is well-suited to obtain global semantic region information while also accommodates local spatial boundary characteristics simultaneously. Experiments on two types of challenging datasets demonstrate that our method outperforms the state-of-the-art approaches on the segmentation of polyps in colonoscopy images and of the optic disc and optic cup in colour fundus images.