Related Experiment Video
Updated: Oct 7, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.0K
Multi-Task Model for Esophageal Lesion Analysis Using Endoscopic Images: Classification with Image Retrieval and
Xiaoyuan Yu1, Suigu Tang1, Chak Fong Cheang1
1Faculty of Information Technology, Macau University of Science and Technology, Taipa, Macau.
Sensors (Basel, Switzerland)
|January 11, 2022
Summary
This study introduces a new deep learning model for analyzing endoscopic images to detect esophageal lesions. The model enhances diagnostic accuracy for both lesion type classification and location segmentation, aiding endoscopists.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Gastroenterology
Background:
- Accurate identification of esophageal lesions during endoscopy is crucial for patient outcomes.
- Current computer-aided diagnosis systems often lack the necessary accuracy and interpretability for clinical adoption.
Purpose of the Study:
- To develop and evaluate a novel multi-task deep learning model for the automatic analysis of endoscopic images.
- To improve the accuracy of esophageal lesion classification and segmentation, providing decision support for endoscopists.
Main Methods:
- A multi-task deep learning framework integrating an image retrieval module for classification confidence and a mutual attention module for segmentation.
- Training and validation on a dataset of 1003 endoscopic images, including esophageal cancer, esophagitis, and normal cases.
Main Results:
- The proposed model achieved a classification accuracy of 96.76% for identifying lesion types.
- The model demonstrated a segmentation performance with a Dice coefficient of 82.47% for lesion localization.
- Experimental results indicate superior performance compared to existing deep learning models.
Conclusions:
- The developed multi-task deep learning model shows significant promise as an effective tool to assist endoscopists in diagnosing esophageal lesions.
- The integration of image retrieval and mutual attention mechanisms enhances diagnostic support capabilities.
- This AI-driven approach can potentially improve the precision and efficiency of esophageal lesion assessment in clinical practice.
