Related Experiment Video
Updated: Aug 5, 2026
04:10
Robotic-assisted Bronchoscopy Combined with Multimodal Imaging for Targeted Lung Cryobiopsies
Published on: July 19, 2024
An Explainable Cryobiopsy AI Model, CRAI, to Predict Progression in Interstitial Pneumonia
Wataru Uegami1, Ethan N Okoshi2, Kris Lami2
1Department of Pathology Informatics, Nagasaki University Graduate School of Biomedical Sciences, Nagasaki, Japan; Department of Pathology, Kameda Medical Center, Kamogawa, Japan.
Summary
An AI model, CRAI, analyzes lung cryobiopsy samples to predict interstitial lung disease (ILD) progression. It accurately identifies progressive cases, aiding early intervention and personalized treatment for better patient outcomes.
Area of Science:
- Pulmonary Medicine
- Artificial Intelligence in Pathology
- Digital Pathology
Background:
- Interstitial lung disease (ILD) presents diverse prognoses and diagnostic challenges due to inter-observer variability in pathological assessments.
- Standardized diagnostic approaches are crucial for accurate prognostication and management of ILD.
- Transbronchial lung cryobiopsy (TBLC) offers a minimally invasive method for obtaining lung tissue for diagnosis.
Purpose of the Study:
- To develop and validate an ensemble artificial intelligence (AI) model, CRAI, for analyzing TBLC specimens.
- To predict disease progression and patient outcomes in ILD using histological features identified by the AI model.
- To enhance the standardization and accuracy of ILD diagnosis and prognostication.
Main Methods:
- Development of a seven-module ensemble AI model (CRAI) to detect 17 significant histological features in TBLC specimens.
- Utilized an XGBoost classifier trained on AI-derived features to predict disease progression.
- Performance evaluation through cross-validation and external testing, assessing respiratory function changes (Δ%FVC) and survival analysis.
Main Results:
- The AI model accurately differentiated between progressive and non-progressive ILD cases in internal cross-validation (135 cases).
- Significant differences in annual Δ%FVC were observed between progressive (-5.198%) and non-progressive (-1.293%) groups, outperforming human pathologists.
- Survival analysis revealed significantly shorter survival times for patients predicted as progressive (p = 0.038).
Conclusions:
- CRAI offers a comprehensive and interpretable AI-driven approach for analyzing TBLC, potentially standardizing ILD diagnosis.
- The model demonstrates significant potential in predicting ILD progression and patient outcomes.
- Early identification of progressive ILD cases by CRAI can guide personalized therapeutic strategies.