Brain MRI Segmentation Binary U-Net Based Architecture using Deep learning algorithm

Abstract

Abstract In medical image processing, Brain tumor segmentation is one of the most important tasks because the segmented region is helpful for the diagnosis and treatment of a disease. In any region, Gliomas may appear and they can be in any shape and size, which is helpful for automatic detection due to segmentation. One of the most evident tools of image processing is to provide complete information about brain tumor anatomy and makes an effective diagnosis through magnetic resonance imaging (MRI). To replace the manual detection system MRI is essential. Kaggle dataset proposed a brain tumor segmentation and detection system. To handle the task of segmentation of glioma, MRI scan based Binary U-Net-based deep learning model is proposed. By using this Binary U-Net Architecture, it identifies the boundary region of the tumor, which is present in the dataset. It detects the type of tumor present in the assigned dataset. This tumor detection is done by masking the original image and binary prediction is done internally by processing through skull removal, feature extraction, and multiple iterations. The binary prediction is compared with the human prediction, if it meets the pixel length, the model is saved and the glioma mask is predicted, else it is returned to data processing. From Binary prediction, some of the parameters are calculated based on the overlap of the segmentation region and pixels between the ground truth and predicted segmentation. Where Sensitivity, Dice, and Sensitivity metrics measure the overlap of pixels and Dice similarity coefficient (DSC) measures the overlap of segmentation region. Compare to previous techniques, Binary U-Net has Recall and Precision rises by 1.07 percent and 1.04 percent once residual dense blocks are included. As a result, residual dense blocks aid in the stabilization of the deep network and the integration of global features.