Related Experiment Video
Updated: Feb 28, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.6K
Simulation-Driven Annotation-Free Deep Learning for Automated Detection and Segmentation of Airway Mucus Plugs on
Lucy Pu1, Naciye Sinem Gezer2, Tong Yu3,4
1Department of Bioengineering, University of Pennsylvania, Philadelphia, PA 19104, USA.
Bioengineering (Basel, Switzerland)
|February 27, 2026
Summary
A novel deep learning framework accurately detects and segments mucus plugs in obstructive lung diseases using synthetic data. This annotation-free approach improves upon manual methods for better quantification of airway mucus plugs.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Medicine
- Pulmonary Medicine
Background:
- Mucus plugs are key indicators in obstructive lung diseases (OLDs) like COPD, asthma, and cystic fibrosis, correlating with lung function and mortality.
- Manual mucus plug identification on CT scans is time-consuming and impractical for large studies, limiting detailed analysis.
- Automated quantification is difficult due to mucus plug heterogeneity, imaging artifacts, and data imbalance.
Purpose of the Study:
- To develop and validate an annotation-free deep learning framework for automated detection and segmentation of airway mucus plugs.
- To overcome challenges in automated mucus plug quantification using a simulation-driven approach.
Main Methods:
- A deep learning framework (nnU-Net) was trained exclusively on synthetically generated mucus plug data.
- Synthetic plugs were created by transferring intensity statistics from adjacent vessels to mimic realistic morphology and texture.
- The synthetic data-trained model (S-Model) was evaluated against a manually annotated model (M-Model) on an independent COPD CT dataset.
Main Results:
- The S-Model demonstrated superior detection performance (0.837 sensitivity) compared to the M-Model (0.757 sensitivity) with fewer false positives (1.91 vs. 3.68 per scan).
- Performance improvements were most significant for medium-to-large mucus plugs (≥6 mm).
- The framework enables scalable, accurate mucus plug quantification without manual voxel-wise annotations on non-contrast chest CT.
Conclusions:
- A simulation-driven, annotation-free deep learning approach can accurately quantify mucus plugs in non-contrast chest CT scans.
- This method offers a scalable solution for analyzing mucus plugs in large cohorts, potentially improving disease burden assessment and treatment response monitoring in OLDs.
- Further validation is needed across diverse populations and imaging protocols to establish generalizability.
