Brain Tumor Segmentation Using Unet

We are at the cusp of a massive Bio-medical revolution. Advances in medical engineering have been supplying us enormous amounts of data, such as medical scans, electroencephalography, genome, and protein sequences. Computer vision algorithms show promise in extracting features and learning patterns from this complex data. One such application is the segmentation of Brain Tumors. There have been a number of somewhat successful attempts at the demarcation of brain tumors through simple Convolutional Neural networks (CNN), CNN-Support Vector Machines, DenseNets, Unets, etc. In this paper, we worked with the database consisting of brain Magnetic Resonance images, with each image composed of three channels together with the manually extracted abnormality masks for segmentation. If implemented for real-world applications, this technology can be used to generate semantic segmentation on Brain MR Images in real-time. We have implemented a U-net architecture, a fully connected CNN. We were successful in demarcating the tumors in the brain MR image accurately.