Brain Tumor Segmentation in MRI Images Using A Modified U-Net Model

Brain tumor segmentation is an essential process to diagnose and monitor the development of cancerous cells in the brain. Conventional segmentation methods rely on experts who manually label radiology individual images. Meanwhile, deep learning has shown tremendous progress in medical image segmentation where minor details are difficult to differentiate. In the paper, we propose a deep learning architecture to automatically segment such radiology images, named UVR-Net model. The proposed architecture is based on the popular U-Net framework which demonstrated its robustness and capabilities in the medical imaging field. Experimental results show that the proposed UVR-Net achieves a Dice score of 0.76, and IOU scores 0.89 compared to the traditional vanilla U-Net architecture by a factor of 11% in terms of Dice score. In addition, we also perform sensitivity analysis for critical parameters and loss functions in the proposed model.