🫁 Introduction¶
ATM26 is part of the SENSAR (Sino-European Surgical Autonomy in Robotics) Network challenges. The SENSAR Network is focused on innovative, reproducible solutions for surgical autonomy and robot assisted interventions. ATM26 is organized in conjunction with MICCAI 2026, which is to be held in Strasbourg, September 27 - October 1, 2026.
Accurate airway tree modeling is clinically important for respiratory disease assessment and endobronchial intervention. Patient-specific airway models can support pre-operative planning, bronchoscopic navigation, and peripheral pulmonary nodule biopsy by helping identify optimal branch-level access routes and reduce procedural uncertainty and operation risks. A complete and topologically reliable airway tree provides the geometric foundation for intervention planning, while anatomical branch labels connect imaging findings with standardized airway nomenclature and clinically interpretable procedure plans.
Building on ATM22 (Airway Tree Modeling): Multi-site, Multi-domain Airway Tree Modeling, which focused on binary airway segmentation, ATM26 extends airway tree modeling from binary segmentation to structured airway understanding, supporting clinically meaningful airway segmentation and branch-wise anatomical labeling. The challenge includes two tracks: binary airway segmentation and branch-wise anatomical labeling.
🏆 Challenge Tracks¶
ATM26 is designed to promote robust, topology-aware, and clinically applicable airway modeling under realistic CT acquisition conditions, including variations in image spacing, image quality, and airway anatomy. The challenge will include two sub-tasks: (i) airway segmentation, which follows the main settings of ATM22 while using high-quality reference masks to support robust benchmarking and continuity with prior work; and (ii) branch-wise anatomical labeling, where participants are required to assign standardized anatomical labels to airway branches and generate structured airway representations. Together, these tasks support key clinical applications such as automated route planning, target localization, and interpretable reporting. Through this design, ATM26 aims to further advance airway modeling methods for endobronchial intervention planning and provide a realistic multi-task benchmark for the community.
Track 1 · Binary Airway Segmentation¶
Track 1 focuses on automatic binary segmentation of the airway tree from chest CT scans. Participants are expected to generate a connected airway mask covering the trachea, main bronchi, lobar bronchi, segmental bronchi, and distal branches.
Unlike conventional segmentation tasks that primarily emphasize voxel-wise overlap, this track encourages methods that preserve airway completeness, connectivity, and distal branch structures. Topological correctness is considered a critical component of clinically useful airway modeling.
Track 2 · Branch-wise Anatomical Labeling¶
Track 2 moves beyond binary segmentation toward structured airway understanding. Participants are required to assign anatomical labels to airway branches and generate an anatomically consistent airway representation.
The target labels include lobar, segmental, and subsegmental airway branches. This task is designed to support automated route planning, lesion-to-airway association, and standardized anatomical reporting for endobronchial intervention.
🚀 How to Submit¶
Submission is not open at the current stage.
The submission system is not available yet at this moment. Participants are first required to complete the Registration for data access. Once approved, participants may download the Dataset and prepare their algorithms in advance.
The submission phase will be opened later. Once it is available, participants can refer to the Submission Guidelines page for more detailed submission requirements and instructions.
📚 Citation¶
If using this dataset, you must cite the papers:
[1] Zhang, M., et al, Multi-site, multi-domain airway tree modeling. Medical Image Analysis, 2023 Dec:90:102957. doi: 10.1016/j.media.2023.102957.
[2] Li, C., Zhang, M., Zhang, C. and Gu, Y., 2025. Reflecting topology consistency and abnormality via learnable attentions for airway labeling. International Journal of Computer Assisted Radiology and Surgery, 20(7), pp.1315-1323.
[3] Zhang, M., Li, C., Xie, F., Liu, Y., Zhang, H., Wu, J., Zhang, C., Yang, J., Sun, J., Yang, G.Z. and Gu, Y., 2024. Airmorph: Topology-preserving deep learning for pulmonary airway analysis. arXiv preprint arXiv:2412.11039.
🎓Challenge Paper¶
We schedule to publish a challenge paper, at most the three authors of the top five performing methods are qualified as authors in the final challenge paper.