Segmentation of Hemorrhagic Stroke Using Graph Convolutional Network

Stroke is a serious global public health problem, and several studies on tools to assist its diagnosis have been conducted, many of them involving image processing techniques and deep learning. However, the use of graph convolutional networks (GCN) for stroke classification and segmentation has yet to be explored. This work uses GCN to implement a hemorrhagic stroke segmentation system in Computed Tomography (CT) scans. The results show that the proposed methodology performs well in segmentation, reaching 95.98% in the Dice coefficient and 92.54% in the Jaccard coefficient.