Dscunet: an Enhanced Brain Tumor Segmentation Model

Deep learning-based techniques have been widely used for brain tumor segmentation in magnetic resonance imaging (MRI) images; however, they are still insufficient for processing small lesions and some tumor boundaries. In this study, we propose a DSCUNet network architecture that combines the COS-SOS attention mechanism and dynamic convolutional blocks to enhance the ability to focus on and extract important features. The network was designed to solve complex problems in medical image segmentation. With a well-designed attention mechanism and convolutional blocks, DSCUNet can focus on information in both spatial and channel dimensions, thus capturing subtle features and the global context of the tumor region more efficiently. The experimental results show that the proposed model achieves significant performance improvement on the BraTS2021 dataset compared to the traditional UNet network and excels in the accuracy, sensitivity, and specificity of tumor segmentation. This study provides a more effective method for brain tumor segmentation, which is expected to provide more accurate assistance for clinical diagnosis and treatment. The experimental results show that for TC, WT, and ET, the mean values of Dice, the main metrics, were <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$91.2 \%, 93.4 \%$</tex>, and 86.9 %, respectively, and the network achieved a higher performance than state-of-the-art segmentation backbone networks.