Uncertainty-Based Dynamic Graph Neighborhoods for Medical Segmentation

In recent years, deep learning based methods have shown success in essential medical image analysis tasks such as segmentation. Post-processing and refining the results of segmentation is a common practice to decrease misclassifications originating from network. addition widely used like Conditional Random Fields (CRFs) which focus on structure segmented volume/area, graph-based approach makes use certain uncertain points graph refines according small convolutional network (GCN). However, there are two drawbacks approach: most edges assigned randomly GCN trained independently To address these issues, we define new neighbor-selection mechanism feature distances combine networks training procedure. According experimental pancreas Computed Tomography (CT) images, demonstrate improvement quantitative measures. Also, examining dynamic neighbors created by our method, between semantically similar parts observed. The proposed method also shows qualitative enhancements maps, demonstrated visual results.