Brain Tumor Segmentation Using 3D ResUNET34

Brain is the most important organ in functioning of a human body. Brain Tumor refers to the growth, development and spread of unwanted dangerous cells in the brain.Brain tumor is a lethal disease and may even lead to death. The motive of this experiment is to integrate the advancement in the field of deep learning with tumor segmentation. For a long period of time brain tumours are segmented manually. This procedure is a hectic and time taking in many complex cases So, automatic accurate segmentation is a need in today’s medical world. Deep learning is used as a tool to address this necessity. In this paper 3D UNET architectures are proposed for Multi Class tumor segmentation. With a conventional 3D UNET as a reference architecture,two modified 3D UNET frameworks are proposed for better performance.The proposed architectures namely are Attention based 3D UNET and 3D ResUNET34, these deep learning models performed well when compared to the baseline reference 3D UNET architecture. 3D ResUNET34 was the best performing model with MeanIoU and Dice coefficient 80.23 and 88.55 respectively.