Brain tumors segmentation using a hybrid filtering with U-Net architecture in multimodal MRI volumes

Brain MRI makes it possible to evaluate brain tumor diagnosis and treatment. There are, however, many challenges for automated brain tumor segmentation, and these challenges have become tougher as deep learning has progressed. Gliomas tumors, which can be proliferated quickly, can develop anywhere and in any shape in the brain. This article presents a new method for automated brain tumor segmentation using hybrid filters and employing 3D medical images. The U-Net model is employed for semantic segmentation. 2D MRIs are taken from 3D MRI, which allows us to view the tumor in two dimensions in three levels (axial, sagittal, and coronal). This ensures that there are not any tumor points that have been overlooked for detection, and the tumor volume images are available in two dimensions. In addition, three types of MRI are used to segment this tumor from each patient. In these three types of MRI, the Glioma tumor is more visible than the fourth type (T2). To improve the tumor segmentation, a hybrid filter, including the bilateral filter and blacktop hat, is used for pre-processing. The intersection of these images is used to evaluate the model. Experimental results showed that the proposed approach demonstrates better tumor segmentation compared to similar studies. In addition, the computational and memory used is lower than what has been done in recent research. The Dice coefficient and Hausdorff distance are estimated at 91.34% and 3.74 for the whole tumor image in the present work, respectively. Experiments show that our method is an optimal method for the segmentation of brain images. © The Author(s), under exclusive licence to Bharati Vidyapeeth's Institute of Computer Applications and Management 2023.