Segmentation and Classification of Brain Tumors using Deep Learning Techniques

Abstract

Brain cancer is classified based on the level of cancerous growth present. There are two major types LGG(Low-Grade Glioma) and HGG (high-grade gliomas). Low grade gliomas are benign tumors (grade I or II). All low-grade gliomas eventually progress to high grade glioma and death. GBM(Glioblastoma) is the most common type of HGG, it is the most aggressive cancer that begins within the brain and is the most common type of malignant brain tumor among adults. It is usually aggressive, which means it can grow fast and spread quickly. Although there is no cure, there are treatments to help ease symptoms. The conventional method for medical resonance brain image classification and tumor detection is by human inspection. There is a need to have a fast and reliable method to classify if the tumor present is cancerous(LGG or HGG). A two staged deep learning architecture is proposed for classification of tumorous images into LGG or HGG. In the first stage we have used an autoencoder regularization for segmenting the input images and stage two uses pre-trained VGG-16 and VGG-19 for final classification. For the evaluation of our model we have used the BraTS 2018 dataset. A final accuracy of 92.54