Multi-Modal Lesion Segmentation Using Deep Convolution Graph-Based Network

The graph-based network aims to increase the visualization of grid-based data into the graph domain. In this paper, a novel fusion architecture of graph con-volutional network (GCN) and features from convolution network based on end to end deep learning for medical lesion segmentation. For multi-modal analysis, features have been captured from each modality of magnetic resonance imaging (MRI). Graph module is learned from neighborhood connectivity with the correlation between modalities and nodes present in the graph. Multi-modal analysis improved segmentation performance as it integrate the independent and individual information from imaging. This improved the segmentation performance as dice similarity coefficient (DSC) was evaluated to 87.4% for the multiple sclerosis (MS) patient datasets.