Transformer-based automated coronary artery segmentation with domain-specific pretraining

   Automated segmentation of coronary arteries in coronary computed tomography angiography plays an important role in the diagnosis and treatment of coronary artery disease. Manual segmentation of coronary arteries requires significant labor costs and is accompanied by subjective errors, which necessitates the development of accurate and reliable automated methods for coronary artery segmentation. The paper presents an approach based on a deep neural network with the Swin-UNETR architecture which combines the advantages of visual transformers and the U-Net structure. To improve the accuracy, a domain-specific transfer learning strategy was used: the model was pre-trained on the ImageCAS dataset, and then further trained on a specialized dataset created for Automated Segmentation of Coronary Arteries (ASOCA) Challenge with expert labeling of coronary arteries. The accuracy of the model was assessed on 10 test Computed Tomography Coronary Angiography cases from the ASOCA dataset. The average Dice coefficient was 0.8778, and the average 95^th percentile Hausdorff distance (HD95) was 11.66 mm. The obtained results demonstrate that the accuracy of the proposed method is at the level of the leading models presented in the official ASOCA Challenge rating and exceeds the average inter-rater labeling. The proposed method provides high accuracy of coronary artery segmentation. In the future, the introduction of post-processing methods such as connected component filtering or vessel tracking, and spatial attention mechanisms can improve the accuracy of arterial contour localization and the adaptability of the model to various types of computed tomography data.

1. Roth G.A., Abate D., Abate K.H., Abay S.M., Abbafati C., Abbasi N., et al. Global, regional, and national age-sex-specific mortality for 282 causes of death in 195 countries and territories, 1980–2017: a systematic analysis for the Global Burden of Disease Study 2017. The Lancet, 2018, vol. 392, no. 10159, pp. 1736–1788. doi: 10.1016/S0140-6736(18)32203-7

2. Montalescot G., Sechtem U., Achenbach S., Andreotti F., Arden C., Budaj A., et al. 2013 ESC guidelines on the management of stable coronary artery disease. European Heart Journal, 2013, vol. 34, no. 38, pp. 2949–3003. doi: 10.1093/eurheartj/eht296

3. Cury R.C., Leipsic J., Abbara S., Achenbach S., Berman D., Bittencourt M., et al. CAD-RADS™ 2.0 - 2022 Coronary Artery Disease-Reporting and Data System. Journal of Cardiovascular Computed Tomography, 2022, vol. 16, no. 6, pp. 536–557. doi: 10.1016/j.jcct.2022.07.002

4. Yang S., Chung J., Lesina K., Doh J., Jegere S., Erglis A., et al. Long-term prognostic implications of CT angiography-derived fractional flow reserve: Results from the DISCOVER-FLOW study. Journal of Cardiovascular Computed Tomography, 2024, vol. 18, no. 3, pp. 251–258. doi: 10.1016/j.jcct.2024.01.016

5. Hurreh M., Knickelbine T., Earls J., Aquino M., Idris A., Cheng V. Enhanced cardiac CTA segmental atherosclerosis detection with artificial intelligence quantitative computed tomography (AIQCT) analysis vs conventional reader interpretation. Journal of Cardiovascular Computed Tomography, 2023, vol. 17, no. 4 suppl., pp. S37-S38. doi: 10.1016/j.jcct.2023.05.093

6. Fedorov A., Beichel R., Kalpathy-Cramer J., Finet J., Fillion-Robin J-C., Pujol S., et al. 3D Slicer as an image computing platform for the Quantitative Imaging Network. Magnetic Resonance Imaging, 2012, vol. 30, no. 9, pp. 1323–1341. doi: 10.1016/j.mri.2012.05.001

7. Yushkevich P.A., Piven J., Hazlett H.C., Smith R.G., Ho S., Gee J.C., Gerig G. User-guided 3D active contour segmentation of anatomical structures: Significantly improved efficiency and reliability. NeuroImage, 2006, vol. 31, no. 3, pp. 1116–1128. doi: 10.1016/j.neuroimage.2006.01.015

8. Dosovitskiy A., Beyer L., Kolesnikov A., Weissenborn D., Zhai X., Unterthiner T., et al. An image is worth 16×16 words: Transformers for image recognition at scale. arXiv, 2020, arXiv:2010.11929. doi: 10.48550/arXiv.2010.11929

9. Fossan F.E., Müller L.O., Sturdy J., Bråten A.T., Jorgensen A., Wiseth R., Hellevik L.R. Machine learning augmented reduced-order models for FFR-prediction. Computer Methods in Applied Mechanics and Engineering, 2021, vol. 384, pp. 113892. doi: 10.1016/j.cma.2021.113892

10. Gharleghi, R., Adikari D., Ellenberger K., Ooi S.Y., Ellis C., Chen C.M., et al. Automated segmentation of normal and diseased coronary arteries — The ASOCA challenge. Computerized Medical Imaging and Graphics, 2022, vol. 97, pp. 102049. doi: 10.1016/j.compmedimag.2022.102049

11. Zeng A., Wu C., Lin G., Xie W., Hong J., Huang M.P., et al. ImageCAS: A large-scale dataset and benchmark for coronary artery segmentation based on computed tomography angiography images. Computerized Medical Imaging and Graphics, 2023, vol. 109, pp. 102287. doi: 10.1016/j.compmedimag.2023.102287

12. Çiçek Ö., Abdulkadir A., Lienkamp S.S., Brox T., Ronneberger O. 3D U-Net: Learning dense volumetric segmentation from sparse annotation. Lecture Notes in Computer Science, 2016, vol. 9901, pp. 424–432. doi: 10.1007/978-3-319-46723-8_49

13. Chen Y.-C., Lin Y.-C., Wang C.-P., Lee C.-Y., Lee W.-J., Wang T.-D., Chen C.-M. Coronary artery segmentation in cardiac CT Angiography using 3D Multi-channel U-net. arXiv, 2019, arXiv:1907.12246. doi: 10.48550/arXiv.1907.12246

14. Liang R., Huang X., Ma J., Ma G., Wang K. 3D U-Net with attention and focal loss for coronary tree segmentation. Research Square. Preprint CC BY 4.0 License, 2021, pp. 1–17. doi: 10.21203/rs.3.rs-717054/v1

15. Wang Q., Xu L., Wang L., Yang X., Sun Y., Yang B., Greenwald S.E. Automatic coronary artery segmentation of CCTA images using UNet with a local contextual transformer. Frontiers in Physiology, 2023, vol. 14, pp. 1138257. doi: 10.3389/fphys.2023.1138257

16. Zhang L., Yin X., Liu X., Liu Z. Medical image segmentation by combining feature enhancement Swin Transformer and UperNet. Scientific Reports, 2025, vol. 15, no. 1, pp. 14565. doi: 10.1038/s41598-025-97779-6

17. Salehi A.W., Khan S., Gupta G., Alabduallah B.I., Almjally A., Alsolai H., et al. A study of CNN and transfer learning in medical imaging: Advantages, challenges, future scope. Sustainability, 2023, vol. 15, no. 7, pp. 5930. doi: 10.3390/su15075930

18. Matsoukas C., Haslum J.F., Sorkhei M., Soderberg M., Smith K. What makes transfer learning work for medical images: Feature reuse & other factors. Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022, pp. 9215–9224. doi: 10.1109/CVPR52688.2022.00901

19. Yosinski J., Clune O., Bengio Y., Lipson H. How transferable are features in deep neural networks? Proc. of the 28^th International Conference on Neural Information Processing Systems, 2014, vol. 2, pp. 3320–3328.

20. Wang X., Zhang X., Wang G., He J., Li Z., Zhu W., et al. OpenMEDLab: An open-source platform for multi-modality foundation models in medicine. arXiv, 2024, arXiv:2402.18028v2. doi: 10.48550/arXiv.2402.18028

21. Tang Y., Yang D., Li W., Roth H.R., Landman B., Xu D., Nath V., Hatamizadeh A. Self-supervised pre-training of Swin Transformers for 3D medical image analysis. Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022, pp. 20698–20708. doi: 10.1109/cvpr52688.2022.02007

22. Yong K.R., Lee S., Uppaluri A., Tan M.O., Ng J.J., Venkatesh S.H., et al. How to report a coronary CT angiogram: Step-by-Step guide for the novice reader. RadioGraphics, 2025, vol. 45, no. 7, pp. e250015. doi: 10.1148/rg.250015

23. Gharleghi R., Adikari D., Ellenberger K., Webster M., Ellis C., Sowmya A., Ooi S., Beier S. Annotated computed tomography coronary angiogram images and associated data of normal and diseased arteries. Scientific Data, 2023, vol. 10, no. 1, pp. 128. doi: 10.1038/s41597-023-02016-2

24. Kirişli H.A., Schaap M., Metz C.T., Dharampal A.S., Meijboom W.B., Papadopoulou S.L., et al. Standardized evaluation framework for evaluating coronary artery stenosis detection, stenosis quantification and lumen segmentation algorithms in computed tomography angiography. Medical Image Analysis, 2013, vol. 17, no. 8, pp. 859–876. doi: 10.1016/j.media.2013.05.007