《计算机应用》唯一官方网站 ›› 2026, Vol. 46 ›› Issue (5): 1667-1676.DOI: 10.11772/j.issn.1001-9081.2025050613
• 前沿与综合应用 • 上一篇
荆莹, 李然(
), 蒋卓, 付子扬, 杜晶颐, 刘琪, 刘吉航
收稿日期:2025-06-13
修回日期:2025-09-16
接受日期:2025-09-25
发布日期:2025-10-17
出版日期:2026-05-10
通讯作者:
李然
作者简介:荆莹(2001—),女,山东临清人,硕士研究生,主要研究方向:深度学习、图像处理基金资助:
Ying JING, Ran LI(
), Zhuo JIANG, Ziyang FU, Jingyi DU, Qi LIU, Jihang LIU
Received:2025-06-13
Revised:2025-09-16
Accepted:2025-09-25
Online:2025-10-17
Published:2026-05-10
Contact:
Ran LI
About author:JING Ying, born in 2001, M. S. candidate. Her research interests include deep learning, image processing.Supported by:摘要:
针对传统SAM(Segment Anything Model)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,提出改进模型ResSAM。该模型引入自动提示编码器消除人工干预的依赖;针对骨干网络进行剪枝优化,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(Squeeze-and-Excitation)与交叉注意力机制降低个体差异和边界模糊的影响,提升模型分割精度。在2个自建数据集Lower Lid和Upper Lid上的实验结果显示,ResSAM的参数量和十亿次浮点运算次数(GFLOPs)指标表现最优;分割结果具有最高Dice值,分别为88.69%和87.75%,以及最高的交并比(IoU)值,分别为79.69%和78.58%。研究结果表明,ResSAM在效率与精度方面均实现了优化,可为睑板腺功能障碍(MGD)的早期预防和临床诊断提供支持。
中图分类号:
荆莹, 李然, 蒋卓, 付子扬, 杜晶颐, 刘琪, 刘吉航. 引入自动提示编码器的SAM睑板腺统一密集分割方法[J]. 计算机应用, 2026, 46(5): 1667-1676.
Ying JING, Ran LI, Zhuo JIANG, Ziyang FU, Jingyi DU, Qi LIU, Jihang LIU. SAM Meibomian gland unified dense segmentation method with introduction of automatic prompt encoder[J]. Journal of Computer Applications, 2026, 46(5): 1667-1676.
| 模型 | Lower Lid | Upper Lid | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Dice/% | IoU/% | Sensitivity/% | Accuracy/% | Time/s | Dice/% | IoU/% | Sensitivity/% | Accuracy/% | Time/s | |
| SAM(10 point) | 47.83 | 31.46 | 73.32 | 92.31 | 20 | 39.24 | 24.66 | 71.29 | 91.12 | 19 |
| SAM(20 point) | 55.92 | 38.79 | 74.66 | 93.86 | 24 | 47.31 | 30.69 | 73.61 | 91.91 | 24 |
| SAM(30 point) | 60.77 | 43.72 | 76.72 | 94.02 | 25 | 54.87 | 37.15 | 74.78 | 92.72 | 26 |
| SAM(40 point) | 60.14 | 48.23 | 77.83 | 94.59 | 31 | 58.16 | 40.19 | 75.92 | 93.60 | 29 |
| SAM(50 point) | 67.95 | 51.48 | 79.25 | 94.93 | 38 | 61.73 | 43.42 | 76.45 | 94.13 | 34 |
| ResSAM | 88.69 | 79.69 | 88.75 | 98.85 | 8.7 | 87.75 | 78.58 | 88.09 | 98.63 | 8.4 |
表1 Lower Lid和Upper Lid数据集中不同点提示数量的SAM与ResSAM的性能对比
Tab. 1 Performance comparison of ResSAM and SAM with different point prompt numbers on Lower Lid and Upper Lid datasets
| 模型 | Lower Lid | Upper Lid | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Dice/% | IoU/% | Sensitivity/% | Accuracy/% | Time/s | Dice/% | IoU/% | Sensitivity/% | Accuracy/% | Time/s | |
| SAM(10 point) | 47.83 | 31.46 | 73.32 | 92.31 | 20 | 39.24 | 24.66 | 71.29 | 91.12 | 19 |
| SAM(20 point) | 55.92 | 38.79 | 74.66 | 93.86 | 24 | 47.31 | 30.69 | 73.61 | 91.91 | 24 |
| SAM(30 point) | 60.77 | 43.72 | 76.72 | 94.02 | 25 | 54.87 | 37.15 | 74.78 | 92.72 | 26 |
| SAM(40 point) | 60.14 | 48.23 | 77.83 | 94.59 | 31 | 58.16 | 40.19 | 75.92 | 93.60 | 29 |
| SAM(50 point) | 67.95 | 51.48 | 79.25 | 94.93 | 38 | 61.73 | 43.42 | 76.45 | 94.13 | 34 |
| ResSAM | 88.69 | 79.69 | 88.75 | 98.85 | 8.7 | 87.75 | 78.58 | 88.09 | 98.63 | 8.4 |
| 17.57 | 33.66 | |||||||
| 16.92 | 32.42 | |||||||
| √ | 63.65 | 82.15 | ||||||
| √ | 61.62 | 80.52 | ||||||
| √ | √ | 75.79 | 88.94 | |||||
| √ | √ | 74.73 | 87.71 | |||||
| √ | √ | √ | 88.69 | |||||
| √ | √ | √ | 87.75 |
表2 ResSAM在Lower Lid和Upper Lid数据集上的消融实验结果 ( %)
Tab. 2 Ablation experimental results of ResSAM model on Lower Lid and Upper Lid datasets
| 17.57 | 33.66 | |||||||
| 16.92 | 32.42 | |||||||
| √ | 63.65 | 82.15 | ||||||
| √ | 61.62 | 80.52 | ||||||
| √ | √ | 75.79 | 88.94 | |||||
| √ | √ | 74.73 | 87.71 | |||||
| √ | √ | √ | 88.69 | |||||
| √ | √ | √ | 87.75 |
| 骨干网络 | Lower Lid | Upper Lid | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Params/106 | GFLOPs | FPS | Dice/% | IoU/% | Params/106 | GFLOPs | FPS | Dice/% | IoU/% | |
| ResNet18 | 11.73 | 24.17 | 92.14 | 80.13 | 71.42 | 11.71 | 24.19 | 92.03 | 79.21 | 67.08 |
| ResNet34 | 21.81 | 44.46 | 78.29 | 83.27 | 75.48 | 21.79 | 44.52 | 78.25 | 82.13 | 71.57 |
| ResNet50 | 25.59 | 52.08 | 70.66 | 85.92 | 79.95 | 25.62 | 52.13 | 70.64 | 85.37 | 74.93 |
| ResNet101 | 44.48 | 95.23 | 52.47 | 91.97 | 87.82 | 44.51 | 95.18 | 52.43 | 91.48 | 87.91 |
| ResDenNet101-Pruned1 | 31.12 | 66.59 | 66.21 | 91.60 | 87.38 | 31.09 | 66.61 | 66.15 | 91.10 | 86.50 |
| ResDenNet101-Pruned2 | 23.71 | 50.82 | 75.37 | 90.83 | 86.71 | 23.68 | 50.79 | 75.32 | 90.28 | 85.85 |
| ResDenNet101-Pruned3 | 18.31 | 38.11 | 85.58 | 88.69 | 79.69 | 18.29 | 38.09 | 85.46 | 87.75 | 78.58 |
表3 不同网络在Lower Lid和Upper Lid数据集上的性能对比
Tab. 3 Performance comparison of different networks on Lower Lid and Upper Lid datasets
| 骨干网络 | Lower Lid | Upper Lid | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Params/106 | GFLOPs | FPS | Dice/% | IoU/% | Params/106 | GFLOPs | FPS | Dice/% | IoU/% | |
| ResNet18 | 11.73 | 24.17 | 92.14 | 80.13 | 71.42 | 11.71 | 24.19 | 92.03 | 79.21 | 67.08 |
| ResNet34 | 21.81 | 44.46 | 78.29 | 83.27 | 75.48 | 21.79 | 44.52 | 78.25 | 82.13 | 71.57 |
| ResNet50 | 25.59 | 52.08 | 70.66 | 85.92 | 79.95 | 25.62 | 52.13 | 70.64 | 85.37 | 74.93 |
| ResNet101 | 44.48 | 95.23 | 52.47 | 91.97 | 87.82 | 44.51 | 95.18 | 52.43 | 91.48 | 87.91 |
| ResDenNet101-Pruned1 | 31.12 | 66.59 | 66.21 | 91.60 | 87.38 | 31.09 | 66.61 | 66.15 | 91.10 | 86.50 |
| ResDenNet101-Pruned2 | 23.71 | 50.82 | 75.37 | 90.83 | 86.71 | 23.68 | 50.79 | 75.32 | 90.28 | 85.85 |
| ResDenNet101-Pruned3 | 18.31 | 38.11 | 85.58 | 88.69 | 79.69 | 18.29 | 38.09 | 85.46 | 87.75 | 78.58 |
表4 Lower Lid数据集上各模型性能对比 ( %)
Tab. 4 Performance comparison of different models on Lower Lid dataset
表5 Upper Lid数据集上各模型性能比较 ( %)
Tab. 5 Performance comparison of different models on Upper Lid dataset
| [1] | 陈远远.基于深度学习的医学图像分割技术研究[D].开封:河南大学,2024. |
| CHEN Y Y. Research on medical image segmentation technology based on deep learning[D]. Kaifeng: Henan University, 2024. | |
| [2] | RONNEBERGER O, FISCHER P, BROX T. U-Net: convolutional networks for biomedical image segmentation[C]// Proceedings of the 2015 Medical Image Computing and Computer-Assisted Intervention, LNCS 9351. Cham: Springer, 2015: 234-241. |
| [3] | BADRINARAYANAN V, KENDALL A, CIPOLLA R. SegNet: a deep convolutional encoder-decoder architecture for image segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(12): 2481-2495. |
| [4] | CHEN L C, PAPANDREOU G, SCHROFF F, et al. Rethinking atrous convolution for semantic image segmentation[EB/OL]. [2025-09-15].. |
| [5] | HE K, ZHANG X, REN S, et al. Deep residual learning for image recognition[C]// Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2016: 770-778. |
| [6] | 曹玉红,徐海,刘荪傲,等.基于深度学习的医学影像分割研究综述[J].计算机应用,2021,41(8):2273-2287. |
| CAO Y H, XU H, LIU S A, et al. Review of deep learning-based medical image segmentation[J]. Journal of Computer Applications, 2021, 41(8): 2273-2287. | |
| [7] | CHHADVA P, GOLDHARDT R, GALOR A. Meibomian gland disease: the role of gland dysfunction in dry eye disease[J]. Ophthalmology, 2017, 124(S11): S20-S26. |
| [8] | YAN X, LI J, LIU Z, et al. Quantitative analysis of morphological and functional features in meibography for Meibomian Gland Dysfunction: diagnosis and grading[J]. Eye and Contact Lens, 2022, 48(1): 32-39. |
| [9] | LIN J W, LIN L J, LU F, et al. Meibomian glands segmentation in infrared images with limited annotation[J]. International Journal of Ophthalmology, 2024, 17(3): 401-407. |
| [10] | GHOSH S, DAS N, DAS I, et al. Understanding deep learning techniques for image segmentation[J]. ACM Computing Surveys, 2020, 52(4): No.73. |
| [11] | AMANO S, SHIMAZAKI J, YOKOI N, et al. Meibomian gland dysfunction clinical practice guidelines[J]. Japanese Journal of Ophthalmology, 2023, 67(4): 448-539. |
| [12] | MARUOKA S, TABUCHI H, NAGASATO D, et al. Deep neural network-based method for detecting obstructive meibomian gland dysfunction with in vivo laser confocal microscopy[J]. Cornea, 2020, 39(6): 720-725. |
| [13] | LAI L, WU Y, FAN J, et al. Automatic meibomian gland segmentation and assessment based on TransUnet with data augmentation[C]// Proceedings of the 2024 International Conference on Intelligent Computing, LNCS 14863. Singapore: Springer, 2024: 154-165. |
| [14] | SHEN N, WANG Z, LI J, et al. Multi-organ segmentation network for abdominal CT images based on spatial attention and deformable convolution[J]. Expert Systems with Applications, 2023, 211: No.118625. |
| [15] | MAZUROWSKI M A, DONG H, GU H, et al. Segment anything model for medical image analysis: an experimental study[J]. Medical Image Analysis, 2023, 89: No.102918. |
| [16] | 孙兴,蔡肖红,李明,等.视觉大模型SAM在医学图像分割中的应用综述[J].计算机工程与应用,2024,60(17):1-16. |
| SUN X, CAI X H, LI M, et al. Review of application of the visual foundation model SAM in medical image segmentation[J]. Computer Engineering and Applications, 2024, 60(17): 1-16. | |
| [17] | XIE B, TANG H, CAI D, et al. Self-Prompt SAM: medical image segmentation via automatic prompt SAM adaptation[EB/OL]. [2025-01-15].. |
| [18] | XIE Z, GUAN B, JIANG W, et al. PA-SAM: prompt adapter SAM for high-quality image segmentation[C]// Proceedings of the 2024 IEEE International Conference on Multimedia and Expo. Piscataway: IEEE, 2024: 1-6. |
| [19] | DONG G, WANG Z, CHEN Y, et al. An efficient segment anything model for the segmentation of medical images[J]. Scientific Reports, 2024, 14: No.19425. |
| [20] | CHENG Z, WEI Q, ZHU H, et al. Unleashing the potential of SAM for medical adaptation via hierarchical decoding[C]// Proceedings of the 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2024: 3511-3522. |
| [21] | LIN T Y, GOYAL P, GIRSHICK R, et al. Focal loss for dense object detection[C]// Proceedings of the 2017 IEEE International Conference on Computer Vision. Piscataway: IEEE, 2017: 2999-3007. |
| [22] | LEE Y, PARK J. CenterMask: real-time anchor-free instance segmentation[C]// Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2020: 13903-13912. |
| [23] | 戚翔宇,苏庆华,张智超,等. 改进YOLOv11的缺陷检测算法[J].计算机科学与应用,2025,15(1):108-116. |
| QI X Y, SU Q H, ZHANG Z C, et al. Improvement of the YOLOv11 defect detection algorithm[J]. Computer Science and Application, 2025, 15(1): 108-116. | |
| [24] | DOSOVITSKIY A, BEYER L, KOLESNIKOV A, et al. An image is worth 16x16 words: Transformers for image recognition at scale[EB/OL]. [2024-12-15].. |
| [25] | HUANG S, LIANG H, WANG Q, et al. SEG-SAM: semantic-guided SAM for unified medical image segmentation[EB/OL]. [2025-08-21].. |
| [26] | CHEN P, XIE L, HUO X, et al. SAM-CP: marrying SAM with composable prompts for versatile segmentation[EB/OL]. [2025-03-11].. |
| [27] | MA J, HE Y, LI F, et al. Segment anything in medical images[J]. Nature Communications, 2024, 15: No.654. |
| [28] | 周涛,刘赟璨,陆惠玲,等.ResNet及其在医学图像处理领域的应用:研究进展与挑战[J].电子与信息学报,2022,44(1):149-167. |
| ZHOU T, LIU Y C, LU H L, et al. ResNet and its applications to medical image processing: research progress and challenges[J]. Journal of Electronics and Information Technology, 2022, 44(1): 149-167. | |
| [29] | ZHU K, HU F, DING Y, et al. A comprehensive review of network pruning based on pruning granularity and pruning time perspectives[J]. Neurocomputing, 2025, 626: No.129382. |
| [30] | PAN P, ZHANG C, SUN J, et al. Multi-scale Conv-Attention U‑Net for medical image segmentation[J]. Scientific Reports, 2025, 15: No.12041. |
| [31] | CHENG B, MISRA I, SCHWING A G, et al. Masked-attention mask transformer for universal image segmentation[C]// Proceedings of the 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2022: 1280-1289. |
| [32] | ZHAO H, SHI J, QI X, et al. Pyramid scene parsing network[C]// Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2017: 6230-6239. |
| [33] | XIE E, WANG W, YU Z, et al. SegFormer: simple and efficient design for semantic segmentation with Transformers[C]// Proceedings of the 35th International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2021: 12077-12090. |
| [1] | 白翔, 李巨川, 王慧民, 景超, 钮键, 张兴忠, 程永强. 基于改进Swin Transformer的电力图像检索方法[J]. 《计算机应用》唯一官方网站, 2026, 46(4): 1334-1343. |
| [2] | 李亚男, 郭梦阳, 邓国军, 陈允峰, 任建吉, 原永亮. 基于多模态融合特征的并分支发动机寿命预测方法[J]. 《计算机应用》唯一官方网站, 2026, 46(1): 305-313. |
| [3] | 许志雄, 李波, 边小勇, 胡其仁. 对抗样本嵌入注意力U型网络的3D医学图像分割[J]. 《计算机应用》唯一官方网站, 2025, 45(9): 3011-3016. |
| [4] | 王芳, 胡静, 张睿, 范文婷. 内容引导下多角度特征融合医学图像分割网络[J]. 《计算机应用》唯一官方网站, 2025, 45(9): 3017-3025. |
| [5] | 张硕, 孙国凯, 庄园, 冯小雨, 王敬之. 面向区块链节点分析的eclipse攻击动态检测方法[J]. 《计算机应用》唯一官方网站, 2025, 45(8): 2428-2436. |
| [6] | 王艺涵, 路翀, 陈忠源. 跨模态文本信息增强的多模态情感分析模型[J]. 《计算机应用》唯一官方网站, 2025, 45(7): 2237-2244. |
| [7] | 姜坤元, 李小霞, 王利, 曹耀丹, 张晓强, 丁楠, 周颖玥. 引入解耦残差自注意力的边界交叉监督语义分割网络[J]. 《计算机应用》唯一官方网站, 2025, 45(4): 1120-1129. |
| [8] | 杨定木, 倪龙强, 梁晶, 邱照原, 张永真, 齐志强. 基于语义相似度的协议转换方法[J]. 《计算机应用》唯一官方网站, 2025, 45(4): 1263-1270. |
| [9] | 令狐鑫瑶, 陈燕, 张鹏程, 刘祎, 桂志国, 赵伟, 董展豪. 基于多尺度引导滤波的宫颈细胞核图像分割[J]. 《计算机应用》唯一官方网站, 2025, 45(4): 1333-1339. |
| [10] | 蒋占军, 李洋, 廉敬, 苗新法. 坐标增强与多源采样的脑肿瘤图像分割[J]. 《计算机应用》唯一官方网站, 2025, 45(3): 996-1002. |
| [11] | 袁宝华, 陈佳璐, 王欢. 融合多尺度语义和双分支并行的医学图像分割网络[J]. 《计算机应用》唯一官方网站, 2025, 45(3): 988-995. |
| [12] | 薛振华, 李强, 黄超. 视觉基础模型驱动的像素级图像异常检测方法[J]. 《计算机应用》唯一官方网站, 2025, 45(3): 823-831. |
| [13] | 邓酩, 徐锦凡, 肖洪祥, 谢晓兰. 改进TransUNet的高效通道注意力医学图像分割网络[J]. 《计算机应用》唯一官方网站, 2025, 45(12): 4037-4044. |
| [14] | 李烨恒, 罗光圣, 苏前敏. 基于改进YOLOv5的Logo检测算法[J]. 《计算机应用》唯一官方网站, 2024, 44(8): 2580-2587. |
| [15] | 姬张建, 杜娜. 基于改进VariFocalNet的微小目标检测[J]. 《计算机应用》唯一官方网站, 2024, 44(7): 2200-2207. |
| 阅读次数 | ||||||
|
全文 |
|
|||||
|
摘要 |
|
|||||