Article Details

Title · Authors · Abstract · Keywords · DOI · References

SAMG: A Meibomian Gland Image Segmentation Model Combining SAM and CNN

ZHANG Chenrui, LI Xiuli

(School of Information Engineering, North China University of Water Resources and Electric Power, Zhengzhou 450046, China)

Section：Artificial Intelligence · Vol. 2025, Issue 6 · Pages：105-109,115 · 2096-4706（2025）06-0105-06 · CLC number：TP391.4 · Document code：A · DOI：10.19850/j.cnki.2096-4706.2025.06.020

Abstract

SAM is a well-known general image segmentation model. Although SAM performs well on natural images, its performance is significantly degraded and its generalization ability is limited when processing meibomian gland medical images with low contrast and blurred boundaries. Therefore, this paper proposes a model SAMG suitable for meibomian gland image segmentation. Based on the image encoder of SAM, the model adds a CNN branch and connects the CNN branch with the original image encoder (ViT branch) through a cross-attention module. In addition, in the final part of the encoder, a SE module enables the model to understand the image content at a finer granularity. This model structure design makes the additional parameters and the amount of calculation relatively small. With the support of professional meibomian gland medical image dataset, comparative experiments show that the SAMG model performs well in meibomian gland image segmentation tasks and can be deployed on entry-level GPU.

Keywords

SAM; CNN; ViT; image segmentation

References

  1. [1] CHAN T C Y，CHOW S S W，WAN K H N，et al. Update on the Association Between Dry Eye Disease and Meibomian Gland Dysfunction [J].Hong Kong Medical Journal，2019，25（1）：38-47.
  2. [2] KIRILLOV A，MINTUN E，RAVI N，et al. Segment Anything [C]//2023 IEEE/CVF International Conference on Computer Vision (ICCV).Paris：IEEE，2023：4015-4026.
  3. [3] ZHANG Y C，JIAO R S. Towards Segment Anything Model (SAM) for Medical Image Segmentation： A Survey [J/OL].arXiv:2305.03678 [eess.IV].[2024-09-10].https://doi.org/10.48550/arXiv.2305.03678.
  4. [4] RONNEBERGER O，FISCHER P，BROX T. U-Net: Convolutional Networks for Biomedical Image Segmentation [C]//Medical Image Computing and Computer-Assisted Intervention(MICCAI 2015).Munich：Springer，2015：234-241.
  5. [5] ZHOU Z W，SIDDIQUEE M M R，TAJBAKHSH N，et al. UNet++: A Nested U-Net Architecture for Medical Image Segmentation [C]//Deep Learning in Medical Image Analysis and Multimodal Learning for Clinical Decision Support.Granada：Springer，2018：3-11.
  6. [6] ISENSEE F，PETERSEN J，KLEIN A，et al. Abstract: nnU-Net: Self-adapting Framework for U-Net-Based Medical Image Segmentation [C]//Bildverarbeitung für die Medizin 2019.Lübeck：Springer，2019:1-22.
  7. [7] CHEN J N，LU Y Y，YU Q H，et al. TransUNet: Transformers Make Strong Encoders for Medical Image Segmentation [J/OL].arXiv:2102.04306 [cs.CV].[2024-09-15].https://doi.org/10.48550/arXiv.2102.04306.
  8. [8] CAO H，WANG Y Y，CHEN J，et al. Swin-Unet: Unet-like Pure Transformer for Medical Image Segmentation [J/OL].arXiv:2105.05537 [eess.IV].[2024-09-08].https://doi.org/10.48550/arXiv.2105.05537.
  9. [9] MA J，HE Y T，LI F F，et al. Segment Anything in Medical Images [J].Nature Communications，2024，15（1）：654[2024-09-20].https://doi.org/10.1038/s41467-024-44824-z.
  10. [10] ZHANG K D，LIU D. Customized Segment Anything Model for Medical Image Segmentation [J/OL].arXiv:2304.13785 [cs.CV].https://doi.org/10.48550/arXiv.2304.13785.
  11. [11] WU J D，JI W，LIU Y P，et al. Medical SAM Adapter: Adapting Segment Anything Model for Medical Image Segmentation [J/OL].arXiv:2304.12620 [cs.CV].[2024-09-13].https://doi.org/10.48550/arXiv.2304.12620.
  12. [12] YU B X，CHANG J L，WANG H X，et al. Visual Tuning [J/OL].arXiv:2305.06061 [cs.CV].[2024-08-30].https://doi.org/10.48550/arXiv.2305.06061.
  13. [13] Xue Lianjian Research on Near Infrared Image Processing Algorithm for Meibomian Glands [D]. Shanghai: Shanghai Jiao Tong University, 2016
  14. [14] LIN X，XIANG Y Y，ZHANG L，et al. SAMUS: Adapting Segment Anything Model for Clinically-Friendly and Generalizable Ultrasound Image Segmentation [J/OL]. arXiv:2309.06824 [cs.CV].[2024-08-25].https://doi.org/10.48550/arXiv.2309.06824.
  15. [15] HU J，SHEN L，SUN G. Squeeze-and-Excitation Networks [J] IEEE Transactions on Pattern Analysis and Machine Intelligence，2018，24（8）：2011-2023.

Funding

Research and Practice Project on Higher Education Teaching Reform in Henan Province in 2024 (2024SJGLX0332); 2023 Henan Province Higher Education Teaching Reform Research and Practice Project Graduate Education Category (2023SJGLX112Y)

About the authors

Zhang Chenrui (2000-), male, Han ethnicity, from Luohe, Henan Province, is currently pursuing a master's degree. His research focuses on medical image segmentation based on deep learning models; Corresponding author: Li Xiuli (1967-), female, Han ethnicity, from Zhengzhou, Henan Province, professor, undergraduate, research direction: computer technology and applications.