Deep Convolutional Neural Network for Multi-class Brain Tumor Classification System in MRI Images

Abstract

Brain tumour is a serious disease which can cause severe damage to the brain cells which eventually turns into a life threatening cancer. The tumour stages when identified early can helps to increase the survival rates of the patients. The performance of the automated brain tumour diagnosis depends on the classification accuracy of the model.. In this article, a deep convolutional neural network (DCNN) is developed for brain tumor classification of brain tumors in MRI images. Specifically, the auto-weight dilated convolutional unit utilized multi-scale convolutional feature maps to acquire brain tumor features at different scales and employed a learnable set of parameters to fuse convolutional feature maps in encoding layers. The AD unit is an effective architecture for feature extraction in the encoding stage. We used the advantages of the U-Net network for deep and shallow features, combined with AD units to multimodal image classification. In this model, the four-channel model inputs correspond to the MRI images of four modes, respectively. The main body of the network is composed of auto-weight dilated (AD) unit, Residual (Res) unit, linear upsampling, and the first and last convolution units.. The network that applied Block-R3 had higher segmentation performance than the networks of Block-R1 and Block-R2. In the U-shaped network, feature extraction at the coding stage is the most important component. Designing the network to extract the features of interest efficiently is crucial. The proposed tumour diagnosis with the optimal feature extraction achieved better results with less time consumption.