Answers are grounded in this paper and cite their sources.
Accurate segmentation of brain tumor from Magnetic Resonance Imaging (MRI) is critical for diagnosis, treatment planning, and surgical intervention. This study proposes a novel deep learning architecture, Dual-Scale Residual Attention U-Net with Swin Transformer Blocks (DSRA-UNet-STB), enhanced with Dense Atrous Spatial Pyramid Pooling (DASPP) for precise brain tumor segmentation. The proposed model integrates EfficientNetB4 as the encoder to extract rich spatial-semantic features, while Swin Transformer modules capture global contextual dependencies. The DASPP module facilitates multiscale feature fusion, improving boundary delineation and the detection of small lesions. The proposed model achieved a Dice score and accuracy of 99.8%, outperforming traditional U-Net, Attention U-Net, and hybrid Convolutional Neural Network (CNN) approaches. It also attained 98.9% sensitivity and 99.2% specificity, indicating strong capability in accurately distinguishing tumor from non-tumor regions. These results demonstrate the potential of DSRA-UNet-STB for high-precision tumor segmentation, making it highly suitable for clinical applications. Future research will focus on real-time deployment and adaptation to multi-modal imaging datasets.
No takes yet. Share an insight, caveat, or question.
J et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: