Related Experiment Video
Updated: Jul 20, 2025
05:22
Author Spotlight: Demonstrating Systematic Endobronchial Ultrasound to New Endoscopists
Published on: August 11, 2023
2.0K
Deep learning with test-time augmentation for radial endobronchial ultrasound image differentiation: a multicentre
Kai-Lun Yu1,2, Yi-Shiuan Tseng3, Han-Ching Yang1
1Department of Internal Medicine, National Taiwan University Hospital Hsin-Chu Branch, Hsinchu, Taiwan.
BMJ Open Respiratory Research
|August 2, 2023
Summary
This study developed an AI model using convolutional neural networks (CNNs) to analyze radial endobronchial ultrasound (rEBUS) images. The AI successfully differentiated malignant from benign lung tumors in rEBUS scans, showing promising diagnostic potential.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Oncology
- Pulmonary Diagnostics
Background:
- Radial endobronchial ultrasound (rEBUS) is crucial for transbronchial biopsies.
- Artificial intelligence (AI) has not yet been applied to analyze rEBUS images.
Purpose of the Study:
- To develop a convolutional neural network (CNN) for distinguishing malignant from benign tumors using rEBUS images.
- To evaluate the AI model's performance in differentiating lung cancer subtypes.
Main Methods:
- Retrospective collection of rEBUS images from multiple medical centers in Taiwan.
- Model training and validation using internal and external datasets, including image augmentation and test-time augmentation (TTA).
Main Results:
- Internal validation showed an Area Under the Curve (AUC) of 0.88.
- External validation AUCs ranged from 0.72 to 0.78, improving to 0.82 after fine-tuning.
- The model demonstrated feasibility in differentiating lung cancer subtypes, with AUCs for adenocarcinoma (0.70) and squamous cell carcinoma (0.64).
Conclusions:
- The developed CNN-based algorithm is feasible for differentiating malignant and benign lesions in rEBUS images.
- AI application in rEBUS analysis shows potential for improving lung cancer diagnosis.
