Abstract

Early diagnosis and selection of an appropriate treatment method will increase the survival of cancer patients. Accurate and reliable brain tumor segmentation is an important component in tumor diagnosis and treatment planning. Glioma is one of the hardest brain tumors in diagnosis because of its irregular shape and blurred borders. Automatic segmentation of glioma brain tumors is a challenging problem due to significant variations in their structure. In this paper, improved UNet-based architectures are presented for automatic segmentation of brain tumors from MRI images. Specifically, we designed the strong Two-Pathway-Residual blocks for UNet structure and proposed three models. Our proposed models’ architectures exploit both local features as well as more global features simultaneously. Furthermore, different from the original UNet, our proposed architectures have fewer parameters. The proposed models were evaluated on the BRATS'2018 database and given good results while the calculation cost was lower than the other methods. DCS, sensitivity, and PPV criteria values used for the segmentation results of the best-proposed model are 89.76%, 89.19%, and 90.65%, respectively.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call