Brain Tumor Segmentation from MRI Images using Hybrid Convolutional Neural Networks

Abstract Brain tumor segmentation is a process of identifying the cancerous brain tissues and labeling them automatically based on the tumor types. Manual segmentation of tumor from brain MRI is time-consuming and error-prone. There is a need for fast and accurate brain tumor segmentation technique. Convolutional Neural Networks (CNNs) have recently shown outstanding performance in computer vision for image segmentation and classification tasks. U-Net, SegNet and ResNet18 are the most popular CNN for image segmentation. The U-Net architecture uses skip connection that captures the fine and soars information but requires higher computational time for training. SegNet is computationally efficient. The ResNet18 also uses skip connection and has a layer which adds inputs from multiple neural network layers to get more accurate results. The proposed U-SegNet and Seg-UNet is a hybridization of the novel architecture, SegNet, and U-Net. The main difference between them is the depth, Seg-UNet uses five convolution blocks compared to U-SegNet, which has three convolution blocks and both the models has a skip connection inspired from U-Net after the first convolutional layer by using a depth concatenation layer. And proposed Res-SegNet is also a hybridization of SegNet and ResNet18. It is inspired by ResNet18 and uses an element-wise addition layer as a skip connection. The SegNet3, SegNet5, U-Net, Seg-UNet, U-SegNet, and Res-SegNet are implemented to compare the performance based on their accuracy. For experimentation, BraTS dataset are used for training and testing the models. From the simulation results, it is observed that the hybrid architecture has higher accuracy.