Brain Tumor Classification With Inception Network Based Deep Learning Model Using Transfer Learning

Abstract

Brain tumor classification is one of the most important aspects in the fields of medical image analysis. As tumors are regarded as precursor to cancers, efficient brain tumor classification can prove life saving. For this reason, Convolutional Neural Network(CNN) based approaches are widely being used for classifying brain tumors. However there lies a dilemma, CNNs are accustomed to large amount of training data for giving better result. It is where transfer learning comes useful. In this paper, we propose 3-class deep learning model for classifying Glioma, Meningioma and Pituitary tumors which are regarded as three prominent types of brain tumor. Our proposed model by adopting the concept of transfer learning uses a pre-trained InceptionV3model extracts features from the brain MRI images and deploys softmax classifier for classification. The proposed system is tested on CE-MRI dataset from figshare and achieves an average classification accuracy of 99%, outperforming all previous methods. Few other performance measures such as precision, recall, F-score are also considered while assessing the performance.