Brain Tumor Segmentation using Deep Learning Techniques: Comparative Analysis

Image segmentation is very important when detecting the presence of tumor in the human brain. Despite classifying Magnetic Resonance Imaging (MRI) images into the presence or absence of a brain tumor, it is also necessary to be able to specify which region in the images is the tumor located. In this paper, we have performed the segmentation of MRI images by making use of different architectures. The architectures tested are the classical Unet, SegNet, LWBNA-Unet, and variant of Unet architectures with pretrained models as a backbone. The models’ performance was evaluated on the testing dataset using intersection over union (IoU) and dice coefficient. The comparative analysis suggested that ResNet5$0_{-}$Unet architecture has shown relatively better performance with an IoU of 0.9863 and a dice coefficient of 0.9924.