Brain Tumor Segmentation with Generative Adversarial Nets

Accurate brain tumor segmentation is the key of clinical diagnostics and treatment planning. However, a large quantity of data produced by Magnetic resonance imaging (MRI) prevents manual segmentation in a reasonable time. So, automatic approaches are required for quick and effective segmentation. Nevertheless, the spatial and structural variability among brain tumors bring a challenge to automatically segment MRI images. In this paper, an automatic end-to-end method based on Generative Adversarial Nets (GAN) is proposed for brain tumor segmentation. This method combines the generating model with the discriminant model and takes GAN instead of conditional random fields (CRF) as high-order smoothing method. The proposed method was validated in the BRATS 2015 database, it can be proven that the proposed method achieves a competitive result and the use of GAN improve the performance of networks. Furthermore, comparing with other recent CNN-based methods, the approach only takes about 10.8s to segment a patient case.