Automatic segmentation of brain tumors from magnetic resonance imaging (MRI) is\na challenging task due to the uneven, irregular and unstructured size and shape of tumors.\nRecently, brain tumor segmentation methods based on the symmetric U-Net architecture have\nachieved favorable performance. Meanwhile, the effectiveness of enhancing local responses for\nfeature extraction and restoration has also been shown in recent works, which may encourage\nthe better performance of the brain tumor segmentation problem. Inspired by this, we try to\nintroduce the attention mechanism into the existing U-Net architecture to explore the effects of\nlocal important responses on this task. More specifically, we propose an end-to-end 2D brain\ntumor segmentation network, i.e., attention residual U-Net (AResU-Net), which simultaneously\nembeds attention mechanism and residual units into U-Net for the further performance improvement\nof brain tumor segmentation. AResU-Net adds a series of attention units among corresponding\ndown-sampling and up-sampling processes, and it adaptively rescales features to effectively enhance\nlocal responses of down-sampling residual features utilized for the feature recovery of the following\nup-sampling process. We extensively evaluate AResU-Net on two MRI brain tumor segmentation\nbenchmarks of BraTS 2017 and BraTS 2018 datasets. Experiment results illustrate that the proposed\nAResU-Net outperforms its baselines and achieves comparable performance with typical brain tumor\nsegmentation methods.
Loading....