Related Experiment Video
Updated: Aug 3, 2026
10:19
Substernal Thyroid Biopsy Using Endobronchial Ultrasound-guided Transbronchial Needle Aspiration
Published on: November 10, 2014
25.6K
Automatic Segmentation of Mediastinal Lymph Nodes and Blood Vessels in Endobronchial Ultrasound (EBUS) Images Using
Øyvind Ervik1,2, Ingrid Tveten3, Erlend Fagertun Hofstad3
1Clinic of Medicine, Nord-Trøndelag Hospital Trust, Levanger Hospital, 7601 Levanger, Norway.
Journal of Imaging
|August 28, 2024
Summary
This study introduces an AI tool for automatically identifying and segmenting mediastinal lymph nodes and blood vessels in endobronchial ultrasound (EBUS) images. The novel U-Net architecture offers a fast and accurate solution for lung cancer staging.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Pulmonology
Background:
- Endobronchial ultrasound (EBUS) is crucial for minimally invasive thoracic lymph node sampling.
- Accurate assessment of mediastinal structures in lung cancer staging is vital but hindered by anatomical variations and subjective image interpretation.
- Current methods face challenges in consistent and precise identification of lymph nodes and vessels in EBUS imaging.
Purpose of the Study:
- To develop and evaluate a novel U-Net architecture-based approach for automatic detection and segmentation of mediastinal lymph nodes and blood vessels in EBUS images.
- To improve the accuracy and efficiency of image analysis in EBUS-guided procedures.
- To provide a tool that assists in lung cancer staging by enhancing the visualization of critical anatomical structures.
Main Methods:
- A novel U-Net architecture was employed for image segmentation.
- A dataset of 1161 EBUS images from 40 patients was annotated for training, validation, and testing.
- Performance was evaluated using Dice similarity coefficient, precision, sensitivity, specificity, and F1 score, alongside processing time.
Main Results:
- The U-Net based approach achieved a Dice similarity coefficient of 0.71 ± 0.35 for lymph nodes and 0.76 ± 0.38 for blood vessels.
- High precision (0.82 ± 0.22 for vessels) and specificity (0.99 ± 0.01 for vessels) were observed.
- The system demonstrated a mean processing and segmentation time of 55 ± 1 ms per image, enabling real-time analysis.
Conclusions:
- The developed U-Net architecture-based approach (EBUS-AI) effectively automates the detection and segmentation of mediastinal lymph nodes and blood vessels in EBUS images.
- The method is accurate, feasible, and rapid, supporting real-time automatic labeling during procedures.
- This AI-driven tool has the potential to enhance the accuracy and efficiency of lung cancer staging and EBUS-guided interventions.
