Multi-Modality Brain Tumor Segmentation Network Based on Collaborative Feature Fusion

Gliomas are among the most deadly tumors. Brain tumor segmentation plays an important role in diagnosing gliomas and improving survival rates. Currently, deep learning has been widely used for brain tumor segmentation and has achieved good segmentation results. Magnetic resonance imaging (MRI) is among the most commonly used techniques for brain tu-mor segmentation. In this paper, we design a network with three encoders to extract information from different MRI modalities and a fusion module based on the attention mechanism so that the network increases the attention to features in different modalities that are useful for segmentation and weakens the effect of useless features. We conduct experiments on two datasets (BraTS2020 and BraTS2021) and have achieved promising results. The segmentation results of our network have improved compared to the classical 3D U-Net on both datasets.