Deep learning approach for detecting and localizing brain tumor from magnetic resonance imaging images

Abstract

Brain is the most important part of the nervous system. Brain tumor is mainly a mass or growth of abnormal tissues in a brain. Early detection of brain tumor can reduce complex treatment process. Magnetic resonance images (MRI) are used to detect brain tumor. In this paper, we have introduced a deep convolutional neural network (CNN) to automatic brain tumor segmentation using MRI medical images which can solve the vanishing gradient problem. Classifying the brain MRI images with Resnet-50 and InceptionV3 in order to identify whether there is tumor or not. After this step, we have compared the accuracy level of both of the CNN models. Thereafter, applied U-Net architecture individually with encoder Resnet-50 and InceptionV3 to avieved promising results. The publicly available low-grade gliomas (LGG) segmentation dataset has been utilized to test the model. Before applying the model on the MRI images preprocessing and several augmentation techniques have been done to obtain quality a dataset. U-net architecture with InceptionV3 provided 99.55% accuracy. On the other hand, our proposed method U-net with encoder ResNet-50 showed 99.77% accuracy.