Abstract: Abstract: The traditional SAM model relies on manual prompts in the segmentation of meibomian gland images, making it difficult to handle issues such as dense glands, irregular shapes, and blurred boundaries. To address this, the ResSAM model is proposed, which eliminates the reliance on manual intervention by introducing an automatic prompt encoder. The backbone network is pruned and optimized to further enhance the model's segmentation efficiency. Focal Loss and Smooth IoU Loss are used for training optimization, and the SE and cross-attention mechanisms are integrated to reduce the impact of individual differences and blurred boundaries, thereby improving segmentation accuracy. Experimental results on two datasets show that ResSAM performs optimally in quantitative evaluations such as Params and FLOPs. The highest Dice values are 88.69% and 87.75%, and the highest IoU values are 79.69% and 78.58%, respectively. The research results indicate that this model achieves optimization in both efficiency and accuracy, providing support for the early prevention and clinical diagnosis of meibomian gland dysfunction.

Key words: ResSAM model, SAM model, meibomian gland image segmentation, automatic prompt encoder, backbone network pruning, SE fusion cross-attention

摘要： 摘 要: 传统SAM模型在睑板腺图像分割中依赖人工提示、难以应对腺体密集、形态不规则及边界模糊的问题。为此提出ResSAM模型，通过引入自动提示编码器消除人工干预的依赖；针对骨干网络剪枝优化，进一步提升模型分割效率；采用Focal Loss和Smooth IoU Loss优化训练，并融合SE与交叉注意力机制以降低个体差异、边界模糊的影响，进而提升分割精度。在两个数据集上训练和评估后的实验结果显示，ResSAM对Params、FLOPs等定量评估，皆表现最优；分割结果拥有最高Dice值分别为88.69%、87.75%，最高的IoU值分别为79.69%、78.58%。研究结果表明，该模型在效率与精度方面均实现优化，为睑板腺功能障碍的早期预防和临床诊断提供了支持。

关键词: ResSAM模型, SAM模型, 睑板腺图像分割, 自动提示编码器, 骨干网络剪枝, SE融合交叉注意力机制