Multimodal Magnetic Resonance Image Brain Tumor Segmentation Based on ACU-Net Network

Medical image segmentation has the significance of research in digital processing. It can locate and identify organ cells, which is essential for clinical analysis, diagnosis, treatment. Since high heterogeneity pathological tissues inconspicuous resolution multimodal magnetic resonance images, we propose a brain tumor method based on ACU-Net network. In beginning, preprocess images to ensure balanced number categories. We adopt deep separable convolutional layers replace ordinary architecture U-Net distinguish spatial correlation appearance mapped channel. introduce residual skip connection into heighten propagation capacity features quicken convergence speed network, realize capture abnormal regions. use active contour model against noise edge cracks, come true tracking deformation solve problem blur edema area, so as divide core enhanced necrotic parenchymal area exactly area. this paper,17926 MRI 335 patients BraTS 2015, 2018, 2019 datasets are used training verifying. Our experiments demonstrate that network better performance than other algorithms subjective vision objective indicators when applied segmentation.