BRAIN TUMOR SEGMENTATION ON FLAIR MR IMAGES WITH U-NET

Abstract

Brain tumors are among the illnesses that, if not treated promptly, can lead to death. It is extremely difficult to detect tumor tissue using only eye examination methods. As a result, Magnetic Resonance (MR) imaging is used to diagnose brain tumors. T1, T1c, T2, and FLAIR MRI sequences provide detailed information about brain tumors. If the segmentation procedure is performed correctly, patients' chances of survival improve. This paper describes an automated brain tumor segmentation for FLAIR sequences in MR images using U-NeT method. The study has been carried out on the BraTS 2018 data set. The models' correctness has been assessed using the binary accuracy, dice coefficient, and IOU assessment criteria. The results of the comparison between the tumor regions identified by the expert physicians and the tumor regions calculated by the U-Net model are as follows: The model has been completed with 99.26% accuracy, and the dice coefficient value, which expresses the similarity on the basis of pixels for the test data, has been found to be 73.99%. Furthermore, the IOU value of 0.59 demonstrated that the model provided accurate estimates for the study.