Partial View Segmentation: A Novel Approach to the Brain Tumor Segmentation
Yida Yin · 2020
This project proposes a novel approach Partial View Segmentation to employ the artificial neural network for the brain tumor segmentation. In contrast to the traditional method, PVS focuses on a part of each multimodal magnetic resonance image scan where brain tumor is most concentrated. The model is trained and assessed based on the public dataset, containing glioblastoma (GBM/HGG) and lower grade glioma (LGG), from Multimodal Brain Tumor Segmentation Challenge 2019. The model eventually achieves 88.15%, 58.51%, 54.11% and 57.11% accuracy for four different labels (background, the GD-enhancing tumor, the peritumoral edema, and the necrotic and non-enhancing tumor core) after 500 epochs. The model with Partial View Segmentation outperforms that with traditional 3D U-net or 2D slices method.