Unsupervised brain tumor segmentation using a symmetric-driven adversarial network

The aim of this study was to computationally model, in an unsupervised manner, a manifold symmetry variations normal brains, such that the learned can be used segment brain tumors from magnetic resonance (MR) images fail exhibit symmetry. An tumor segmentation method, named as symmetric driven generative adversarial network (SD-GAN), proposed. SD-GAN model trained learn non-linear mapping between left and right images, thus being able present variability (symmetry) brains. then reconstruct brains based on higher reconstruction errors arising their unsymmetrical. evaluated two public benchmark datasets (Multi-modal Brain Tumor Image Segmentation (BRATS) 2012 2018). provided best performance with accuracy superior state-of-the-art methods performed comparably (<3% lower Dice score) supervised U-Net (the most widely method for medical images). This demonstrated features presenting (i.e., inherent anatomical variations) modelled using unannotated MR segmenting tumors.