Classification of MR Brain Images for Detection of Tumor with Transfer Learning from Pre-trained CNN Models

Magnetic resonance imaging is one of the non-invasive imaging techniques widely employed to diagnose brain diseases. Early diagnosis and treatment of brain tumors is essential. It is a time consuming process for the radiologists to manually classify MR brain images into normal and images with tumors. In this paper, an automated method based on Convolutional Neural Network (CNN) is proposed for detection of tumor in brain images. The CNN model pre-trained on the huge image database of ImageNet, is used to train the input brain images. The high level features extracted are given as input to the fully connected layer followed by softmax activation. The method is tested on MR brain images from database of Harvard medical school. An analysis is done with utilization of three pre-trained models - VGG16, ResNet and Inception. It is able to achieve an accuracy of 100% on the experimented database. Also, from the results, it is inferred that data augmentation improves classification accuracy.