A Deep Dive Into Brain Tumor Segmentation Using U-Net: Harnessing The Power Of Fully Convolutional Networks

Abstract

Brain cancer is a deadly illness that has a profoundly detrimental impact on the lives of those who are affected. Consequently, early brain tumour detection boosts patient survival rates and enhances the effectiveness of treatments. On the other hand, early brain tumour detection is a difficult undertaking and an unmet need. Magnetic Resonance Imaging (MRI) is a noninvasive imaging method, it is typically the first step in the diagnosis and segmentation of brain tumours. Thus, during the past several years, a lot of research has been done on creating reliable automated methods for detecting brain tumours. Segmenting an MRI brain tumour is crucial for automated brain tumour detection and subsequent analysis. This research article shall focus on developing a deep learning architecture for performing segmentation of tumor from the brain MRI. Also using the same architecture models will be developed to segment necrosis, edema, and enhancing tumor region from the MRI images. The fully convolutional architecture (U-Net) based models were trained, tested, and validated for performing segmentation. This shall assist the physician in planning further diagnosis and treatment process.