Brain Tumour Image Segmentation using Deep Networks
DOI:
https://doi.org/10.17762/msea.v71i4.1116Abstract
Automated segmentation of brain tumours from multimodal MR images is a key part of figuring out how a disease is progressing and keeping track of it. Since gliomas are cancerous and different from one another, efficient and accurate segmentation techniques are used to divide tumours into classes within the tumour. In tasks of semantic segmentation, deep learning algorithms do better than the more traditional, context-based computer vision approaches. Convolutional Neural Networks have made a big difference in the accuracy of brain tumour segmentation. They are used a lot for biomedical image segmentation. In this paper, we propose a group of two segmentation networks, a 3D CNN and a U-Net, that work together in a simple but important way to make predictions that are more accurate. Both models were trained separately on the BraTS-19 challenge dataset and evaluated to produce segmentation maps that were very different from each other in terms of how they divided up tumour sub-regions and how they were put together to make the final prediction.