Automatic Localization and Classification of Brain Tumors using Deep Residual Networks and MRI Images

Automatic segmentation tools can serve as valuable aids for healthcare professionals providing additional information for clinical decision–making. This can include assessing disease severity, predicting outcomes, and tailoring treatment strategy. Indeed, brain tumor segmentation poses challenges in tumor shapes, size, and locations, as well as the complex and intricate nature of brain anatomy. Traditional segmentation methods may struggle with theses complexities. Deep neural networks, particularly those employing residual blocks help mitigate the vanishing gradient issue, allowing an effective training of deep networks, by enabling the learning of complex features by aiding the network in capturing subtle details crucial for accurate segmentation. This paper proposes a reliable encoder –decoder neural network based on U-net architecture and residual blocks performing the multiclass segmentation of brain tumors. The experimental study will be carried on BRATS 2020 dataset representing different tumour shapes, locations, sizes and image intensities.