Related Experiment Video
Updated: Apr 9, 2026
Unilateral Lung Volume Analysis Using Micro-CT for Enhanced Assessment of Pulmonary Fibrosis in Preclinical Models
Published on: June 20, 2025
Machine Learning Identification of Progressive Pulmonary Fibrosis in ILD Using KL-6 and Routine Blood Parameters.
Yifan Chen1,2,3, Qianyue Yang1,2,3, Xingyi Liang4
1Department of Clinical Laboratory, State Key Laboratory of Respiratory Disease, National Center for Respiratory Medicine, National Clinical Research Center for Respiratory Disease, Guangzhou Institute of Respiratory Health, The First Affiliated Hospital of Guangzhou Medical University, Guangzhou, China.
Machine learning models using routine blood tests and KL-6 can diagnose progressive pulmonary fibrosis (PPF) in interstitial lung disease (ILD) noninvasively. This approach aids early detection in resource-limited settings, improving patient outcomes for this high-mortality condition.
Area of Science:
- Pulmonary Medicine
- Medical Informatics
- Biomarker Discovery
Background:
- Progressive pulmonary fibrosis (PPF) is a severe form of interstitial lung disease (ILD) with high mortality.
- Diagnosing PPF-ILD is challenging in resource-limited settings due to the lack of advanced imaging and need for invasive procedures.
- Early and accurate diagnosis of PPF-ILD is crucial for timely intervention and improved patient outcomes.
Purpose of the Study:
- To develop and validate a machine learning model for diagnosing PPF-ILD.
- To utilize routine blood parameters and the Krebs von den Lungen-6 (KL-6) biomarker for diagnosis.
- To create a noninvasive and clinically applicable diagnostic tool for resource-constrained environments.
Main Methods:
- A dataset of 10,687 ILD patients was used, divided into training and temporal validation cohorts.
- Machine learning algorithms, including Lasso + random forest (RF) and glmBoost + RF, were employed.
- Model performance was evaluated using area under the curve (AUC), calibration, and decision curve analysis (DCA).
Main Results:
- The Lasso + RF model achieved an AUC of 0.998 (training) and 0.842 (validation).
- The glmBoost + RF model (10 variables) showed an AUC of 0.996 (training), 0.831 (validation), 90.0% sensitivity, 61.0% specificity, and 83.3% F1 score.
- KL-6 emerged as the strongest predictor (OR = 6.20), with both models demonstrating excellent calibration and DCA net benefit.
Conclusions:
- A streamlined machine learning model using routine blood parameters and KL-6 offers a noninvasive method for PPF-ILD diagnosis.
- This model provides objective and clinically applicable performance comparable to more complex methods.
- The developed tool is suitable for early PPF-ILD detection in resource-limited settings, addressing a critical unmet need.
