Related Experiment Video
Updated: Jul 2, 2025
04:10
Author Spotlight: Expanding Interventional Pulmonology Research with Robotic-Assisted Bronchoscopy
Published on: July 19, 2024
611
Machine-Learning-Based Classification Model to Address Diagnostic Challenges in Transbronchial Lung Biopsy.
Hisao Sano1,2,3, Ethan N Okoshi1, Yuri Tachibana1,3
1Department of Pathology Informatics, Nagasaki University Graduate School of Biomedical Sciences, Nagasaki 852-8588, Nagasaki, Japan.
Cancers
|February 24, 2024
Summary
A new machine learning model can help differentiate benign from malignant transbronchial biopsy (TBLB) lung nodule samples. This AI tool aids diagnosis, potentially reducing patient need for repeat procedures.
Area of Science:
- Pulmonary Pathology
- Computational Pathology
- Oncology Diagnostics
Background:
- Distinguishing benign from malignant pulmonary nodules via transbronchial lung biopsy (TBLB) is diagnostically challenging.
- Accurate differentiation is crucial for appropriate patient management and treatment planning.
Purpose of the Study:
- To develop and validate a machine learning (ML) classifier for TBLB specimens.
- To improve the diagnostic accuracy of TBLB in classifying pulmonary nodules.
Main Methods:
- Six potential histologic markers were assessed by pathologists: interface bronchitis/bronchiolitis (IB/B), plasma cell infiltration (PLC), eosinophil infiltration (Eo), lymphoid aggregation (Ly), fibroelastosis (FE), and organizing pneumonia (OP).
- A gradient-boosted decision-tree (XGBoost) ML model was trained on 200 TBLB cases and tested on 51 cases with known benign or malignant outcomes.
Main Results:
- Five markers showed associations with benign conditions (AUC 0.58–0.75), with IB/B being the strongest predictor.
- Fibroelastosis (FE) was the sole indicator associated with malignancy (AUC = 0.66).
- The XGBoost model achieved an AUC of 0.78 for classifying benign vs. malignant TBLB specimens on the test set.
Conclusions:
- The developed ML model shows promise in distinguishing benign from malignant TBLB samples, independent of tumor cell presence.
- This AI-driven approach can enhance diagnostic accuracy and potentially decrease the need for repeat sampling procedures.
