Related Experiment Video
Updated: May 20, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.6K
Attention-Enhanced Multi-Task Deep Learning Model for Classification and Segmentation of Esophageal Lesions
Muhammad Aftab1,2,3, Faisal Mehmood4, Kashif Iqbal Sahibzada5,6
1Pathophysiology Department, School of Basic Medical Sciences, Zhengzhou University, Zhengzhou 450001, China.
ACS Omega
|March 24, 2025
Summary
This study presents a new deep learning model for detecting esophageal lesions, improving early diagnosis of esophageal cancer. The model aids endoscopists by classifying and segmenting lesions, enhancing diagnostic accuracy.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Gastroenterology
Background:
- Accurate detection and segmentation of esophageal lesions are critical for gastrointestinal disease diagnosis and treatment.
- Early esophageal cancer detection is challenging, impacting patient survival rates.
- Existing diagnostic tools require enhancement for improved precision and efficiency.
Purpose of the Study:
- To introduce a novel multitask deep learning model for automatic diagnosis of esophageal lesions.
- To integrate classification and segmentation tasks to assist endoscopists.
- To improve the accuracy and efficiency of esophageal lesion detection and segmentation.
Main Methods:
- Developed a multitask deep learning model using MobileNetV2 architecture with a mutual attention module.
- Integrated classification and segmentation tasks for comprehensive lesion analysis.
- Evaluated the model on Early Esophageal Cancer (EEC), CVC-ClinicDB, and KVASIR datasets.
Main Results:
- Achieved high classification accuracies: 98.72% (CVC-ClinicDB), 98.95% (KVASIR), and 99.12% (EEC).
- Obtained excellent F1-scores: 98.08% (CVC-ClinicDB), 98.32% (KVASIR), and 99.00% (EEC).
- Demonstrated strong segmentation performance with a Dice coefficient of 92.73% and IoU of 91.54%.
Conclusions:
- The proposed multitask deep learning model effectively assists endoscopists in evaluating esophageal lesions.
- The model enhances diagnostic precision and alleviates endoscopist workload.
- This approach shows significant improvement over state-of-the-art models for esophageal lesion analysis.
