Segmenting Brain Tumors from MRI Using Cascaded Multi-modal U-Nets

Gliomas are the most common primary brain tumors, and their accurate manual delineation is a time- consuming and very user-dependent process. Therefore, developing automated techniques for reproducible detection and segmentation of brain tumors from magnetic resonance imaging is a vital research topic. In this paper, we present a deep learning-powered approach for brain tumor segmentation which exploits multiple magnetic-resonance modalities and processes them in two cascaded stages. In both stages, we use multi-modal fully-convolutional neural nets inspired by U-Nets. The first stage detects regions of interests, whereas the second stage performs the multi-class classification. Our experimental study, performed over the newest release of the BraTS dataset (BraTS 2018) showed that our method delivers accurate brain-tumor delineation and offers very fast processing—the total time required to segment one study using our approach amounts to around 18 s.