ABSTRACT
In this research, we investigated tumor segmentation in brain MRI images using the deep convolutional network. The data set used in this research is BRATS 2020 images. Despite the high ability of deep learning methods in learning hierarchical features and automatically extracting features in segmentation, there are problems such as low generalization power. But considering that the feature extraction stage is done automatically, the feature extraction speed is much higher than in the case where the feature extraction is done manually. First, preprocessing operations are performed to increase the data quality for use in the learning phase on MRI images. Then the proposed model is used for brain tumor segmentation. In the proposed model, by using blocks with different filter sizes in the U-Net structure, in addition to local features, global features in MRI images are used simultaneously.
