Brain Tumor Segmentation in FLAIR MRI Images Using Deep Learning Techniques

Brain tumor segmentation represents a challenging task in medical images processing. Brain lesions can cause disruptions in different functions of the brain depending on their location and size. Accurate and rapid identification of brain lesions is a crucial stage of the treatment process. Brain tumor segmentation aims to establish a precise delineation of brain tumor regions using accurately located masks. Manual segmentation has proven to be ineffective with a large dataset in terms of efficiency, time consumption, cost, and prone to the radiologists’ errors. Recently, researchers have used deep learning techniques to segment the tumor from MRI images in a fully automatic way and without human intervention. In this research, an automated tumor segmentation method based on U-Net-based 2D CNN has been introduced in detail. This method uses less time and memory and is easier to be constructed. Datasets collected from LGG and HGG patients were analyzed. The fully automated segmentation results were compared to manually delineated ground truth images showing promising results with dice score coefficient, where the similarity of the whole tumor parts is about 0.903 in both LGG and HGG datasets