MMTransNet: Multimodality MRI brain tumor segmentation with transformer

초록

In MRI brain tumor segmentation task, it is typical for researchers to combine all modalities and serve as an input into a segmentation model for feature extraction, thus ignore the complex relationship between these modalities. In this work, we propose a multimodality encoders-decoder network with transformer block (MMTransNet) to extract information from four different modalities (T1, T1-c, T2, FLAIR) of the MICCAI BraTS2018 dataset for segmenting brain tumors. Experimental result shows that our MMTransNet shows higher performance than previous brain tumor segmentation methods.