MRI (magnetic resonance image) analysis is crucial for diagnosis, monitoring, and treatment of brain tumors. Manual MRI segmentation of brain tumors requires professional knowledge and costs huge amount of time. Automatic segmentation for multimodal 3D MR images is thus very desirable for clinical applications. In this paper, we present an automatic brain tumor segmentation method based on the U-Net architecture, which is composed of ROI (region of interest) extraction and 3D segmentation networks with long and short skip connections. The introduced method is able to capture detailed features and improves the problem of the imbalanced classification of tumors' sub-regions. The method was evaluated on the BRATS 2017 dataset and the results were promising.

3D CNN, Brain tumor segmentation, ROI extraction
dx.doi.org/10.1109/ROBIO49542.2019.8961648
2019 IEEE International Conference on Robotics and Biomimetics, ROBIO 2019
Department of Systems and Computer Engineering

Huang, J. (Jing), Zheng, M. (Minhua), & Liu, P. (2019). Automatic brain tumor segmentation using 3D architecture based on ROI extraction. In IEEE International Conference on Robotics and Biomimetics, ROBIO 2019 (pp. 36–40). doi:10.1109/ROBIO49542.2019.8961648