Adapting Segmentation Foundation Models to 3D Pulmonary Vasculature: A Translational Research Agenda

Segmentation foundation models promise reusable representations, broad task coverage, and reduced dependence on task-specific training, but pulmonary vasculature exposes a difficult domain gap. Chest computed tomography is volumetric, often anisotropic, and visually dominated by structures larger than the peripheral vessels of interest. Pulmonary vessels form a sparse branching network whose clinical value depends on connectivity, caliber, and anatomical identity rather than foreground overlap alone. This research agenda examines how foundation models can be adapted to this setting without discarding their pretrained knowledge or incurring the full cost of high-resolution three-dimensional attention. MorVess is used as a concrete case because it freezes a two-dimensional Segment Anything Model encoder, adds a lightweight 2.5D adapter, predicts mask, distance, and thickness fields, and applies global-local fusion with staged optimization. The design illustrates a productive compromise between general representation and specialized geometry, while its limited depth context and sensitivity to anisotropic voxels reveal open problems. The agenda proposes research priorities in volumetric context, parameter-efficient adaptation, anatomy-aware prompting, annotation design, cross-domain evaluation, uncertainty, and clinical workflow integration. It argues that successful translation requires a model to preserve vascular structure across scanners and diseases, produce inspectable uncertainty, and support downstream measurements or corrections. Foundation-model adaptation should therefore be evaluated as a complete evidence and workflow problem, not only as a competition for higher Dice.

1. Mao, F., Chen, Y., Wu, B., Lin, L., Dai, J., Li, Z., ... & Qin, F. (2026). MorVess: Morphology-Aware Pulmonary Vessel Segmentation Network. arXiv preprint arXiv:2606.24214.
2. Chu, Y., Luo, G., Zhou, L., Cao, S., Ma, G., Meng, X., ... & Gao, X. (2025). Deep learning-driven pulmonary artery and vein segmentation reveals demography-associated vasculature anatomical differences. Nature Communications, 16, 2262.
3. Cicek, O., Abdulkadir, A., Lienkamp, S. S., Brox, T., & Ronneberger, O. (2016). 3D U-Net: Learning dense volumetric segmentation from sparse annotation. In Medical Image Computing and Computer-Assisted Intervention - MICCAI 2016 (pp. 424-432). Springer.
4. Eppink, J., Kervadec, H., van Capelleveen, J., Verhoeff, J., Senan, S., & Bohoudi, O. (2026). PaSAL: A deep learning pipeline for pulmonary artery-vein segmentation and anatomical labeling in thoracic CT. In Proceedings of the 9th International Conference on Medical Imaging with Deep Learning (pp. 3561-3592). PMLR.
5. Hatamizadeh, A., Nath, V., Tang, Y., Yang, D., Roth, H. R., & Xu, D. (2022). Swin UNETR: Swin transformers for semantic segmentation of brain tumors in MRI images. In Brainlesion: Glioma, Multiple Sclerosis, Stroke and Traumatic Brain Injuries (pp. 272-284). Springer.
6. He, K., Chen, X., Xie, S., Li, Y., Dollar, P., & Girshick, R. (2022). Masked autoencoders are scalable vision learners. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (pp. 16000-16009).
7. Hu, E. J., Shen, Y., Wallis, P., Allen-Zhu, Z., Li, Y., Wang, S., Wang, L., & Chen, W. (2022). LoRA: Low-rank adaptation of large language models. In International Conference on Learning Representations.
8. Isensee, F., Jaeger, P. F., Kohl, S. A. A., Petersen, J., & Maier-Hein, K. H. (2021). nnU-Net: A self-configuring method for deep learning-based biomedical image segmentation. Nature Methods, 18(2), 203-211.
9. Kirillov, A., Mintun, E., Ravi, N., Mao, H., Rolland, C., Gustafson, L., ... & Girshick, R. (2023). Segment Anything. In Proceedings of the IEEE/CVF International Conference on Computer Vision (pp. 4015-4026).
10. Ma, J., He, Y., Li, F., Han, L., You, C., Wang, B., ... & Wang, B. (2024). Segment anything in medical images. Nature Communications, 15, 654.
11. Ronneberger, O., Fischer, P., & Brox, T. (2015). U-Net: Convolutional networks for biomedical image segmentation. In Medical Image Computing and Computer-Assisted Intervention - MICCAI 2015 (pp. 234-241). Springer.
12. Shit, S., Paetzold, J. C., Sekuboyina, A., Ezhov, I., Unger, A., Zhylka, A., ... & Menze, B. H. (2021). clDice - A novel topology-preserving loss function for tubular structure segmentation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (pp. 16560-16569).
13. Wang, H., Guo, S., Ye, J., Deng, Z., Cheng, J., Li, T., ... & He, J. (2025). SAM-Med3D: A vision foundation model for general-purpose segmentation on volumetric medical images. IEEE Transactions on Neural Networks and Learning Systems, 36(10), 17599-17612.