Brain Tumor Segmentation Using U-Net

Brain tumor is an abnormal growth of cells within the brain or the central nervous system. It can be classified as primary or metastatic. One of the most prevalent methods for detecting brain tumors is by conducting a Magnetic Resonance Imaging (MRI). However, as different radiologist may have different perspective, the use of state-of-the-art models in brain tumor segmentation such as U-Net architecture are preferred. With more than 160.000 people living with brain and other nervous system cancer in the United States, the timely detection of these tumors are vital. In this paper, three models: Regular U-Net, Upgraded U-net, and Attention U-Net are proposed to segment the brain tumor and compared using Dice Score Coefficient (DSC). Between the three models, the regular U-Net yields the best DSC of 0.3902, 0.6877, and 0,6534 for the Necrotic, Edema, and Enhancing Tumor mask respectively. While these results are promising, they also highlight areas for potential improvement. Further research could explore alternative data pre-processing methods or adjustments to model architecture to enhance segmentation accuracy. The findings of this study emphasize the potential of advanced deep learning models like UNet to revolutionize brain tumor detection, and also the addition and comparison of various models and added techniques that can further refine each of the model's performance, offering more consistent and accurate results, which are crucial for timely and effective patient care.