Brain Tumor Segmentation and Tumor Prediction Using 2D-VNet Deep Learning Architecture

Abstract

Segmentation of brain tumors is a difficult task because of the enormous variation in the intensity and size of gliomas. The tumor type Glioma is the highly prevalent malignant tumor in the brain, with a high death rate and a morbidity rate of more than 3%. In the clinic, MRI is the most common way of detecting brain cancers. Automatic segmentation is difficult because of the overlapping area between the intensity distributions of healthy, enhancing, non-enhancing and edema regions. Segmenting brain tumour areas utilising multi-modal MRI scan pictures can help with treatment observation, post-diagnosis monitoring, and patient impacts evaluation. Manual segmentation, on the other hand, is still the most common procedure in clinical brain tumour segmentation, which takes time and results in significant performance variations across operators. For this reason, the development of accurate and consistent automatic segmentation technology is required. Convolutional neural networks (CNNs), have shown promise in brain tumor segmentation due to their powerful learning capacity. This article suggests an 2D-VNet model for brain tumor segmentation and enhancing the prediction. The presented model was successfully segmented brain tumors and predict the result in enhancing tumor and real enhancing tumor. Experiment with BRATS2020 benchmarks dataset, we found that Loss (for Training: .0025, Testing: .0032 and Validation: .0031), Dice Coefficient (for Training: .9974, Testing: .9967 and Validation: .9968) and Accuracy (for Training: .9971 Testing: .9963 and Validation: .9964).