Deep learning-based detection and classification of airway nodules on CT for early medical intervention
Abstract
Objectives: Airway lesions are common CT findings in interval and post-screen lung cancers; however, airway nodules are frequently overlooked. Robust artificial intelligence (AI) tools for characterizing airway nodules remain limited. This study aimed to develop and evaluate deep learning-based models for the detection and classification of airway nodules on low-dose CT(LDCT) as a second-reading tool.
Materials and methods: 1813 segmental and more proximal airway nodules were retrospectively identified from 1107 LDCT of 1043 participants. A sphere-based center-points matching network, first used to establish a lung nodule detection model, was transferred to airway nodules and enhanced by incorporating coordinate and multiscale attention. The model was trained on 907 proximal airway nodules and independently tested on 738 proximal nodules across four hospitals. The clinical utility of the AI tool was evaluated through human-AI interaction. For nodule classification, a 3D SEResNeXt101-based model was trained, validated and tested using 286 airway nodules (121 benign, 165 malignant) with a 6:2:2 split and evaluated in an independent testing dataset.
Results: On the independent testing set, the detection model achieved a high sensitivity of 73.7%. Human-AI interaction demonstrated AI assistance significantly improved radiologists’ performance, increasing sensitivity from 39.5% to 72.9% and F1-score from 49.4% to 77.2%, while reducing diagnosis time from 83.8 s to 61.8 s. The classification model showed high sensitivities of 82.1%, 71.4%, 94.3%, 72.7% and specificities of 62.7%, 73.9%, 73.9%, 87.5% across multicenter training, validation, testing and independent testing sets, respectively.
Conclusion: This AI system significantly enhances the detection and classification of proximal airway nodules, guiding early medical intervention.
Key Points: Question Improved early detection and classification of airway nodules are critical for lung cancer mortality reduction, but AI-based CAD systems targeting airway nodules remain scarce.
Findings The DL models for detection and classification of proximal airway nodules on chest CT were developed and assisted radiologists to achieve higher detection accuracy.
Critical Relevance Statement The clinical applications of the proposed DL-based system for detecting and classifying proximal airway nodules on chest CT scans help radiologists reduce missed/delayed diagnoses and shorten diagnostic time as a second-reading tool, guiding medical intervention and improving clinical outcomes.
Related
- Multi-view attention-based deep learning for benign–malignant classification of pulmonary nodules
- Machine learning-based association analysis and cross-sectional classification of physical exercise behaviors and mental health among vocational medical students
- A deep learning-based silhouette gait detection framework for fall risk classification in Parkinson’s disease
- Real-Time AI-Based ACR TI-RADS Classification of Thyroid Nodules on the Ultrasound Console: A Prospective Agreement Study in a Region with Historical Iodine Deficiency
- Normalized risk-based evaluation of machine learning–based classification models: A multiclass approach with applications in medical devices
BibTeX
@article{liu2026deep,
title = {{Deep learning-based detection and classification of airway nodules on CT for early medical intervention}},
author = {Shulin Liu and Ping Yi and Yaheng Fan and Jing Hu and Chao Zheng and Peng Lei and Sai Li and Fangxue Yang and Chenyu Li and Yangzhen Hou and Yuxin Li and Guanghua Luo and Xiangjun Fang and Haijun Zheng and Harrison Bai and Hui Zhou and Weihua Liao},
journal = {Insights into Imaging},
year = {2026},
month = nov,
doi = {10.1186/s13244-026-02446-7},
url = {https://link.springer.com/article/10.1186/s13244-026-02446-7},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Deep learning-based detection and classification of airway nodules on CT for early medical intervention
AU - Shulin Liu
AU - Ping Yi
AU - Yaheng Fan
AU - Jing Hu
AU - Chao Zheng
AU - Peng Lei
AU - Sai Li
AU - Fangxue Yang
AU - Chenyu Li
AU - Yangzhen Hou
AU - Yuxin Li
AU - Guanghua Luo
AU - Xiangjun Fang
AU - Haijun Zheng
AU - Harrison Bai
AU - Hui Zhou
AU - Weihua Liao
PY - 2026
DA - 2026/11/05
JO - Insights into Imaging
DO - 10.1186/s13244-026-02446-7
UR - https://link.springer.com/article/10.1186/s13244-026-02446-7
AB - Objectives: Airway lesions are common CT findings in interval and post-screen lung cancers; however, airway nodules are frequently overlooked. Robust artificial intelligence (AI) tools for characterizing airway nodules remain limited. This study aimed to develop and evaluate deep learning-based models for the detection and classification of airway nodules on low-dose CT(LDCT) as a second-reading tool. Materials and methods: 1813 segmental and more proximal airway nodules were retrospectively identified from 1107 LDCT of 1043 participants. A sphere-based center-points matching network, first used to establish a lung nodule detection model, was transferred to airway nodules and enhanced by incorporating coordinate and multiscale attention. The model was trained on 907 proximal airway nodules and independently tested on 738 proximal nodules across four hospitals. The clinical utility of the AI tool was evaluated through human-AI interaction. For nodule classification, a 3D SEResNeXt101-based model was trained, validated and tested using 286 airway nodules (121 benign, 165 malignant) with a 6:2:2 split and evaluated in an independent testing dataset. Results: On the independent testing set, the detection model achieved a high sensitivity of 73.7%. Human-AI interaction demonstrated AI assistance significantly improved radiologists’ performance, increasing sensitivity from 39.5% to 72.9% and F1-score from 49.4% to 77.2%, while reducing diagnosis time from 83.8 s to 61.8 s. The classification model showed high sensitivities of 82.1%, 71.4%, 94.3%, 72.7% and specificities of 62.7%, 73.9%, 73.9%, 87.5% across multicenter training, validation, testing and independent testing sets, respectively. Conclusion: This AI system significantly enhances the detection and classification of proximal airway nodules, guiding early medical intervention. Key Points: Question Improved early detection and classification of airway nodules are critical for lung cancer mortality reduction, but AI-based CAD systems targeting airway nodules remain scarce. Findings The DL models for detection and classification of proximal airway nodules on chest CT were developed and assisted radiologists to achieve higher detection accuracy. Critical Relevance Statement The clinical applications of the proposed DL-based system for detecting and classifying proximal airway nodules on chest CT scans help radiologists reduce missed/delayed diagnoses and shorten diagnostic time as a second-reading tool, guiding medical intervention and improving clinical outcomes.
SN - 1869-4101
ER -