Article Details

Abstract

Automated Diagnostic Systems (ADS) have shown great potential for the early detection of polyps during colonoscopy, thereby reducing the incidence of colorectal cancer. However, the high cost of annotation and strict privacy constraints make it difficult to obtain high-quality endoscopic images, which hampers ADS development. To determine plausible locations and shapes of polyps and improve the realism and diversity of synthetic images, we fine-tune the Stable Diffusion inpainting model with LoRA and integrate Kolmogorov–Arnold Network (KAN) modules. This yields realistic augmented data. Experimental results on the LDPolypVideo dataset show that our model achieves an FID of 11.532 and an IS of 2.901, marking a significant improvement over conventional methods.

Keywords

polyp image generation; Stable Diffusion inpainting model; LoRA fine-tuning; Kolmogorov–Arnold Network

References

1. [1] Siegel R L，Miller K D，Fuchs H E，et al. Cancer statistics, 2022 [J].CA: A Cancer Journal for Clinicians，2022，72（1）：7-33.
2. [2] Bray F，Jacques F，Isabelle S，et al. Global Cancer Statistics 2018: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries [J].CA: a Cancer Journal for Clinicians，2018，68（6）：394-424.
3. [3] American Cancer Society. Colorectal Cancer Early Detection, Diagnosis, and Staging [EB/OL].[2025-12-17].https://www.cancer.org/cancer/colon-rectal-cancer/detectiondiagnosis-staging/detection.html.
4. [4] Chen Z，Li Y W，Xing X H，et al. Medical Federated Learning with Joint Graph Purification for Noisy Label Learning [J/OL].Medical Image Analysis，2023，90：102976.https://doi.org/10.1016/j.media.2023.102976.
5. [5] Fan D P，Ji G P，Zhou T，et al. PraNet: Parallel Reverse Attention Network for Polyp Segmentation [C]//Medical Image Computing and Computer Assisted Intervention - MICCAI 2020，2020：263-273.
6. [6] Li C X，Liu H Y，Liu Y F，et al. Endora: Video Generation Models as Endoscopy Simulators [C]//Medical Image Computing and Computer Assisted Intervention - MICCAI 2024，2024：230-240.
7. [7] Fagereng J A，Thambawita V L B，Storås A，et al. PolypConnect: Image Inpainting for Generating Realistic Gastrointestinal Tract Images with Polyps [C]//2022 IEEE 35th International Symposium on Computer-Based Medical Systems (CBMS)，2022：66-71.
8. [8] Macháček R，Mozaffari L，Sepasdar Z，et al. Mask-conditioned latent diffusion for generating gastrointestinal polyp images [C]//ICDAR' 23: Proceedings of the 4th ACM Workshop on Intelligent Cross-Data Analysis and Retrieval，2023：1-9.
9. [9] Dorjsembe Z，Pao H K，Xiao F. Polyp-DDPM: Diffusion-Based Semantic Polyp Synthesis for Enhanced Segmentation [J/OL].arXiv:2402.04031 [cs.CV].https://doi.org/10.48550/arXiv.2402.04031.
10. [10] Thambawita V，Salehi P，Sheshkal S A，et al. SinGAN-Seg: Synthetic Training Data Generation for Medical Image Segmentation [J/OL].arXiv:2107.00471 [eess.IV].https://doi.org/10.48550/arXiv.2107.00471.
11. [11] Song H，Shin Y. Semantic Polyp Generation for Improving Polyp Segmentation Performance [J].Journal of Medical and Biological Engineering，2024，44（2）：280-292.
12. [12] Qadir H A，Balasingham I，Shin Y. Simple U-net Based Synthetic Polyp Image Generation: Polyp to Negative and Negative to Polyp [J/OL].Biomedical Signal Processing and Control，2022，74：103491.https://doi.org/10.1016/j.bspc.2022.103491.
13. [13] Sharma V，Kumar A，Jha D，et al. ControlPolypNet: Towards Controlled Colon Polyp Synthesis for Improved Polyp Segmentation [C]//2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)，2024：2325-2334.
14. [14] Ma J J，Lu F Q，Huang S L，et al. S. Generalize Polyp Segmentation Via Inpainting Across Diverse Backgrounds and Pseudo-Mask Refinement [J/OL].arXiv:2405.12784 [cs.CV].https://doi.org/10.48550/arXiv.2405.12784.
15. [15] Liu S Y，Chen Z，Yang Q S，et al. Polyp-Gen: Realistic and Diverse Polyp Image Generation for Endoscopic Dataset Expansion [C]//2025 IEEE International Conference on Robotics and Automation (ICRA)，2025：15776-15782.
16. [16] Hu E J，Shen Y L，Wallis P，et al. LoRA: Low-Rank Adaptation of Large Language Models [J/OL].arXiv:2106.09685 [cs.CL].https://doi.org/10.48550/arXiv.2106.09685.
17. [17] Liu Z M，Wang Y X，Vaidya S，et al. KAN: Kolmogorov-Arnold Networks [J/OL].arXiv:2404.19756 [cs.LG].https://doi.org/10.48550/arXiv.2404.19756.
18. [18] Rombach R，Blattmann A，Lorenz D，et al. High-Resolution Image Synthesis with Latent Diffusion Models [C]//2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)，2022：10674-10685.
19. [19] Ronneberger O，Fischer P，Brox T. U-Net: Convolutional Networks for Biomedical Image Segmentation [J/OL].arXiv:1505.04597 [cs.CV].https://doi.org/10.48550/arXiv.1505.04597.
20. [20] Li G J，Liu T Y，Majeed A P P A，et al. GroupKAN: Efficient Kolmogorov-Arnold Networks via Grouped Spline Modeling [J/OL].arXiv:2511.05477 [cs.CV].https://doi.org/10.48550/arXiv.2511.05477.
21. [21] Ma Y R，Che X J，Cheng K，et al. LDPolypVideo Benchmark: A Large-Scale Colonoscopy Video Dataset of Diverse Polyps[C]//Medical Image Computing and Computer Assisted Intervention - MICCAI 2021，2021：387-396.
22. [22] Heusel M，Ramsauer H，Unterthiner T，et al. GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium [J/OL].arXiv:1706.08500 [cs.LG].https://doi.org/10.48550/arXiv.1706.08500.
23. [23] Salimans T，Goodfellow I J，Zaremba W，et al. Improved Techniques for Training GANs [C]//Advances in Neural Information Processing Systems 29，2016：1-9.
24. [24] Avrahami O，Fried O，Lischinski D. Blended Latent Diffusion [J].Transactions on Graphics，2023，42（4）：1-11.

About the authors

Ou Qikun (2002-), male, Han ethnicity, from Hengyang, Hunan, is a master's student with research interests in image segmentation and image denoising.