Research Article - Biomedical Research (2019) Volume 30, Issue 6
Effective High-grade gliomas segmentation based on H-Dense U-Net with CNN architecture.
In the medical field, brain tumor identification and segmentation is still a challenging task. Due to its complex structure, the region segmentation of the tumor from medical images is a strenuous process. Existing techniques are failed to predict the tumor cell from the brain in an accurate manner. To overcome this hindrance, we develop an H-Dense U-Net CNN architecture model fused with a pattern based learning strategy that ensures a better performance criterion. In every field of research, CNN plays a vital role and improvised system performance. The hybrid densely connected U-Net contains a 2D Dense U-Net which extracts the intra-slice features and 3D counterpart accumulating volumetric context. The intra-slice representation and inter-slice features are merged through a hybrid feature fusion layer (HFF). H-Dense U-Net also called an end to end system. We utilize the HGG Tumour imaging data from BRATS 2018 dataset where four different modalities are available (T1, T1c, T2, and FLAIR). Among the different modalities, we preferred FLAIR images. It is subject to preprocessing and then the segmentation process. After segmentation, the tumor region gets masked. Thus the tumor is detected and the system measures the area occupied by the tumor through region props. The key contribution of the paper is to segment the tumor accurately from the healthy tissue and provide better visualization for the physicians. The accuracy of our approach is about 98.74%. Hence our proposed system provides high accuracy and efficient value in segmenting the brain tumors. It also promotes the segmentation performance to the next level in the medical field.Author(s): Seetha J, S Selvakumar Raja