Automatic Brain Tumor Segmentation from MRI Images Using Variants of U-Net Model

Segmenting brain tumor is an essential task in order to diagnosis, planning, and monitoring of patients with brain tumors. U-Net, a convolutional neural network technique, has recently received a lot of attention for its performance in semantic segmentation tasks such as brain tumor segmentation. Despite this, many U-Net-based approaches have been presented, and as they all use different architectural modifications and training strategies, it’s important to compare and evaluate how well they do. This research provides an extensive evaluation of the various U-Net-based methods currently used for brain tumors segmentation. To investigate the potential of advanced U-Net variants for precise brain tumor segmentation, this study compared the performance of U-Net, Attention U-Net, U-Net++, U-Net 3+, and ELU-Net using the BraTS 2018 dataset. In this research, the Dice coefficient, IoU score, and accuracy are used to measures the performance of all the models. We show experimentally that all U-Net-based approaches perform well in brain tumor segmentation. However, the various U-Net types show significantly varying levels of performance. We also compare the computation complexity of each approach. Furthermore, we also emphasize the potential of the development and clinical implementation of each U-Net-based method and explore their respective strengths and limitations. The study of U-Net-based methods for brain tumor segmentation compares their performance & characteristics that can provide important insights. The results can help researchers and clinicians to choose the best U-Net variants for their particular requirements, leading to better accuracy in brain tumor segmentation and better care and treatment outcomes for patients.