Classification and Detection of Brain Tumors from Magnetic Resonance Imaging Scans using Deep Transfer-Learning

Abstract

Detection and classification of tumors in brain using MRIs is a quite strenuous errand when done manually. Automation of these processes becomes an important task to fulfill. Magnetic Resonance Images (MRI) are widely used for medical imaging as the they provide an excellent image quality for tumors, tissues and other parts inside the body. Early interception or detection of tumors in the brain can play a significant role in treatment. Further, classifying the type of tumor based on their location in the brain can assist surgeons to treat the affected patient efficiently. Deep learning techniques are extensively applied for extraction of features and classification of images into different classes. Identification and classification of medical images like MRI scans is commonly done using CNN or Convolutional Neural Network architectures in deep learning. This paper proposes a transfer learning approach based on different pre-trained CNN architectures such as VGG-19 and ResNet101. A custom hybrid model based on a pre-trained Inception-Resnet-v2 with attributes of both Inception and ResNet architectures is proposed. Comparative analysis of all the models were done with accuracy and AUC as the evaluation metrics. The highest accuracy was procured by the Inception-Resnet-v2 based hybrid model, which came up to be 99.30%.