Related Experiment Video
Updated: Jul 11, 2025
04:35
Author Spotlight: Advancing Biopsy Techniques with Transesophageal Ultrasound
Published on: November 21, 2023
478
A deep learning-based system for mediastinum station localization in linear EUS (with video).
Liwen Yao1,2,3,4, Chenxia Zhang2,3,4, Bo Xu1
1Department of Gastroenterology, Wuhan Fourth Hospital, Wuhan, Hubei Province, China.
Endoscopic Ultrasound
|November 16, 2023
Summary
This study introduces the EUS-MPS system for real-time mediastinal EUS station recognition, improving trainee accuracy and simplifying image interpretation for mediastinal diseases.
Area of Science:
- Medical Imaging
- Gastroenterology
- Artificial Intelligence
Background:
- Endoscopic Ultrasound (EUS) is vital for diagnosing mediastinal diseases.
- Efficient mediastinal EUS imaging relies on accurate station identification.
- Current methods require significant expertise in anatomy and technical skills.
Purpose of the Study:
- To develop and validate a real-time system for mediastinal EUS station recognition.
- To reduce the technical difficulty of mediastinal ultrasound image interpretation.
- To aid in standardizing mediastinal EUS scanning procedures.
Main Methods:
- A deep learning model, EUS-MPS, was trained on 33,010 mediastinal EUS images.
- Validation included image, video, and external datasets from multiple hospitals.
- A man-machine contest and crossover study assessed system performance and impact on trainee accuracy.
Main Results:
- The EUS-MPS system achieved high accuracy in image (90.49%) and video (83.80%) validation.
- External validation demonstrated 89.85% accuracy.
- Trainee accuracy in station recognition significantly improved by 13.26% in a crossover study.
Conclusions:
- The deep learning-based EUS-MPS system demonstrates strong performance in mediastinal station localization.
- The system has the potential to shorten the learning curve for EUS examinations.
- It can contribute to establishing standardized mediastinal scanning protocols.
