Brain Tumor Segmentation From MRI Images Using Deep Learning

Abstract

Brain tumor is an abnormal growth of tissues inside the skull. Not all tumors are cancerous but still it affects to the nervous system. Early detection of a tumor can in- crease the chance of survival. That's why it is more important to identify tumor more accurately. Manual detection needs highly understanding about the tumor and experi- ence and mainly it is highly dependant on human perspective. With the help of deep learning it is possible to develop a model which can identify brain tumor in a very early stage. Model can be trained through a large number of MRI images which helps to make it more accurate. Convolutional neural network is used to analysis visual images. It is useful to extract different features from given image and classify into different groups. U-Net architecture is implemented using CNN and specially designed for bio medical im- age segmentation. U-Net uses high number of parameters which leads to the over-fitting. It is also computationally intensive task. To overcome this issue, inception network is introduced inside the U-Net architecture. Using it, complete deep neural network can be formed to detect brain tumor. Glioma is a malignant type of tumor which directly affects nervous system. Current model is based on detection of glioma tumor. Training im- ages and ground truth images are provided under Brain Tumor Segmentation Challenge. Data-sets are divided into HGG (High Grade Glioma) & LGG (Low Grade Glioma). Both are used to train model as well as test it.