ATM22 Airway Tree Modeling
Thoracic CT dataset developed to support automated airway tree segmentation and modeling research.
2026-09-1593726133
70
Overview
Schema Version
https://atlas.rsna.org/schemas/2025-11/dataset.json
Name
ATM22 Airway Tree Modeling
Link
https://atm22.grand-challenge.org/
Indexing
Keywords: Pulmonary Airway Segmentation, Airway Tree Modeling, Deep Learning, Image Segmentation, Topological Analysis, Computed Tomography, COVID-19, Ground-glass opacity, Consolidation
Content: CT, CH, RS
Author(s)
Minghui Zhang
Yangqian Wu
Hanxiao Zhang
Yulei Qin
Hao Zheng
Wen Tang
Corey Arnold
Chenhao Pei
Pengxin Yu
Yang Nan
Guang Yang
Simon Walsh
Dominic C. Marshall
Matthieu Komorowski
Puyang Wang
Dazhou Guo
Dakai Jin
Ya’nan Wu
Shuiqing Zhao
Runsheng Chang
Boyu Zhang
Xing Lv
Abdul Qayyum
Moona Mazher
Qi Su
Yonghuang Wu
Ying’ao Liu
Yufei Zhu
Jiancheng Yang
Ashkan Pakzad
Bojidar Rangelov
Raul San Jose Estepar
Carlos Cano Espinosa
Jiayuan Sun
Guang-Zhong Yang
Yun Gu
Organization(s)
Institute of Medical Robotics, Shanghai Jiao Tong University
Shanghai Chest Hospital
InferVision Medical Technology Co., Ltd.
Imperial College London
Alibaba DAMO Academy
Contact
gzyang@sjtu.edu.cn
Comments
A public benchmark for pulmonary airway segmentation, ATM’22 provides 500 large-scale CT scans with detailed pulmonary airway annotations. The dataset was collected from multiple sites and includes a portion of noisy CT scans from patients with COVID-19, featuring ground-glass opacity and consolidation, to test algorithm robustness and generalization.
Date
Published: 2023-03-10
Dataset
Motivation
To provide a public benchmark for the quantitative comparison of pulmonary airway segmentation algorithms, particularly deep learning-based approaches. The dataset aims to address the lack of large-scale, publicly annotated data, which hinders the development and evaluation of new methods for resolving finer details of distal airways for early intervention of pulmonary diseases.
Partitioning scheme
The dataset consists of 500 CT scans, partitioned into 300 for training, 50 for validation, and 150 for testing.
Noise
The dataset includes a portion of CT scans from patients with COVID-19, which are considered a 'noisy domain'. These scans introduce bias attributes such as bilaterally scattered irregular patches of ground glass opacity, thickening of inter-lobular or intra-lobular septa, and consolidation.