基于Swin Transformer与注意力机制的腺样体图像分割模型OA

To address the specific challenges of adenoid regions,such as low contrast be-tween lesions and surrounding tissues,unclear edge textures,and occlusions caused by na-sal cavity obstructions,a U-shaped adenoid segmentation model was proposed based on Swin Transformer and self-attention mechanisms.The Swin Transformer was employed as the encoder to efficiently extract adenoid features.An attention-based feature fusion module was introduced to effectively integrate low-level detailed information and high-level semantic information.Additionally,a multi-level aggregation module was introduced to enhance feature complementarity through a cross-scale information fusion strategy.Ex-periments conducted on an internal adenoid dataset demonstrate that the proposed method achieves significant improvements in mIoU,mDice,Accuracy,Precision and Recall,vali-dating its effectiveness in improving adenoid segmentation performance.