 This paper examines the various improvements made to the UNET architecture in order to improve its performance when it comes to segmenting brain tumors from MRI scans. The authors compare four different UNET architectures, including 3D UNET, ATTENTION UNET, R2 ATTENTION UNET, and modified 3D UNET on the BRATS 2020 dataset. They also evaluate each model's performance using metrics such as dye score and Haasdorf distance 95 percent, and discuss the limitations and challenges of medical image analysis. This article was authored by Ramayusef, Shaker Khan, Gaurav Gupta, and others.