Comparative Study on Architecture of Deep Neural Networks for Segmentation of Brain Tumor using Magnetic Resonance Images

Abstract

The state-of-the-art works for the segmentation of brain tumor using the images acquired by Magnetic Resonance Imaging (MRI) with their performances are analyzed in this comparative study. First, the architectures of convolutional neural networks (CNN) and the variants of U-shaped Network (U-Net), a kind of Deep Neural Network (DNN) are compared and their differences are highlighted. The publicly available datasets of MRI images specifically Brain Tumor Segmentation (BraTS) are also discussed. Next, the performances of tumor segmentation of various methods in the literature are compared using the parameters such as Dice score and Hausdroff distance (95). This study concludes that the U-Net based architectures using the BraTS-2019 dataset outperform well compared with other CNN based architectures.

Keywords

• Brain tumor
• convolutional neural networks
• deep neural networks
• image segmentation
• magnetic resonance imaging
• U-Net