An Efficient Cascaded CNN Architecture for Brain Tumor Detection in MRI Images

Abstract

This research work proposed an automated tumor detection system based on cascaded Convolutional Neural Network (CNN) architecture. In this, each input has convolved separately with three kernels (3 x 3, 5 x 5 and 7 x 7) and their three output feature maps are cascaded to be processed into the hierarchy of two convolution and pooling layers followed by fully connected (FC) layer. In FC layer, the softmax classification technique has performed to find the pixel-wise classification and to detect whether the particular image consisting of tumor or not. This proposed work is tested with BRATS-2018 dataset of both Low-Grade Gliomas (LGG) and High-Grade Gliomas (HGG) brain images. Further, this work has evaluated using different metrics namely accuracy, precision, recall, F1-score, specificity and sensitivity. Thus, this method outperforms well with 96% accuracy, 98% precision, 98% F1-score and 99% sensitivity, demonstrating that the tumor identification has achieved 5% better accuracy than the existing tumor detection methods