Brain Tumor Segmentation with an Auxiliary Network for Hybrid Transformer-CNN Architecture

Brain tumor segmentation plays a crucial role in medical image analysis and faces multiple challenges. This study proposes a novel brain tumor segmentation network—Hybrid Feature Network. This network combines Convolutional Neural Networks and Transformer architectures to improve segmentation accuracy of multi-modal MRI data. ANTCNet designs an auxiliary feature extraction network to analyze the correlations between different modalities, and enhances local feature extraction capabilities through the SwinMBConvBlock, reducing computational complexity and better adapting to 3D data. The feature fusion and recalibration modules effectively bridge the semantic gap between Transformer and CNN features, solving the problems of fuzzy and inaccurately aligned boundaries. This method performs exceptionally well in the segmentation tasks of enhancing tumor regions, tumor cores, and the entire tumor.