Brain Tumour Segmentation from 3D MRI using U-net

Brain cancers, despite their rarity, are extremely lethal. Among them, the most frequently seen dominant brain tumours are histological gliomas. In addition to diagnostic variables and heterogeneous histological sub-regions that include four tumour regions (peritumoural edema, necrotic core, enhancing and non-enhancing tumour cores), gliomas are highly invasive because they grow quickly and can infect the central nervous system. Magnetic Resonance Imaging (MRI) can be used for monitoring, analyzing, and treatment planning of tumours. To visualize the brain, the MRI uses four different modalities: T1-weighted, T2-weighted, T1ce-weighted, and flair. Though these modalities give complementary information about tumour core regions, they can be used together for monitoring and analyzing sub-regions of tumour cores. Segmentation is a technique that provides information about tumour – affected boundaries. Several tumours have irregular shapes, sizes, and ambiguous boundaries, making it difficult for physicians to manually segment the tumour. That is why the need for an automated segmentation system occurs. Taking into account the need, we present a system, tested and evaluated on BRATS 2020 (training + validation) dataset to automatically segment brain tumour, thereby eliminating human errors. The U-NET model, which is a fully convolutional neural network, is used to perform the segmentation task. Following a successful execution, our system generates segmented regions of necrotic, edema, and enhancing tumour core.