Related Experiment Video
Updated: Aug 5, 2026
05:22
Systematic Endobronchial Ultrasound - The Six Landmarks Approach
Published on: August 11, 2023
Artificial Intelligence and Endobronchial Ultrasound for Lymph-Node Detection and Characterization: A Pioneer
Miguel Mascarenhas1,2,3, Gema Diaz Nuevo4, Sara Lopes5
1Department of Community Medicine, Information and Health Decision Sciences, Faculty of Medicine, University of Porto, 4200-319 Porto, Portugal.
Journal of Clinical Medicine
|July 28, 2026
Summary
An AI model using Vision Transformers can automatically detect and classify lymph nodes in endoscopic ultrasound images, improving lung cancer staging accuracy and reducing variability in this minimally invasive procedure.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- Accurate mediastinal and hilar lymph node (LN) assessment is crucial for lung cancer staging and treatment.
- Endobronchial ultrasound-guided transbronchial needle aspiration (EBUS-TBNA) is the standard for mediastinal staging but is operator-dependent and variable.
- A Vision Transformer-based AI model was developed for automated LN detection and classification on EBUS images.
Purpose of the Study:
- To develop and evaluate an AI model for automatic lymph node detection and classification using EBUS images.
- To assess the performance of the AI model in identifying benign versus malignant lymph nodes.
- To explore the potential of AI to enhance diagnostic accuracy and reduce interobserver variability in EBUS-TBNA procedures.
Main Methods:
- A multicenter, retrospective study utilizing EBUS images from three referral centers.
- Development of a Vision Transformer-based AI model trained on annotated EBUS images with bounding boxes for LN localization.
- Performance evaluation using mean average precision (mAP50) for detection and accuracy, sensitivity, specificity, and AUROC for characterization.
Main Results:
- The AI model achieved a mAP50 of 0.79 for lymph node detection in the test set.
- For malignant lymph node characterization, the model demonstrated 90.6% accuracy, 85.4% sensitivity, and an AUROC of 0.938.
- The dataset comprised 6492 images from 27 patients, with 16 malignant lymph nodes identified.
Conclusions:
- The study demonstrates the feasibility of AI-assisted EBUS interpretation for nodal staging workflows.
- AI-assisted EBUS has the potential to improve diagnostic performance and reduce operator variability in lung cancer staging.
- Larger prospective studies are recommended prior to clinical implementation of the AI model.
