Related Experiment Video
Updated: Sep 7, 2025
08:05
Lung CT Segmentation to Identify Consolidations and Ground Glass Areas for Quantitative Assesment of SARS-CoV Pneumonia
Published on: December 19, 2020
14.3K
Chest L-Transformer: Local Features With Position Attention for Weakly Supervised Chest Radiograph Segmentation and
Hong Gu1, Hongyu Wang1, Pan Qin1
1Faculty of Electronic Information and Electrical Engineering, Dalian University of Technology, Dalian, China.
Frontiers in Medicine
|June 20, 2022
Summary
A new weakly supervised model, Chest L-Transformer, improves chest radiograph segmentation by focusing on local features and lesion position dependencies. This approach overcomes limitations of global models, enhancing diagnostic accuracy for thoracic diseases.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer Vision
Background:
- Chest radiographs are vital for diagnosing thoracic diseases.
- Weakly supervised deep learning is popular for medical image segmentation.
- Existing models struggle with chest radiograph symmetry and lesion-position dependencies.
Purpose of the Study:
- To develop a weakly supervised model for chest radiograph segmentation that addresses global symmetry and lesion-position dependencies.
- To improve segmentation accuracy in medical imaging using novel deep learning techniques.
Main Methods:
- Proposed Chest L-Transformer, a weakly supervised model utilizing local features and Transformer attention.
- Chest L-Transformer models dependencies between lesions and their positions, focusing on disease-prone areas.
- Log-Sum-Exp voting unifies pixel-level and image-level predictions for training.
Main Results:
- Significant improvement in segmentation performance compared to state-of-the-art methods.
- Maintained competitive classification performance.
- Successfully addressed misclassification issues caused by global symmetry in chest radiographs.
Conclusions:
- Chest L-Transformer offers a superior approach to weakly supervised segmentation of chest radiographs.
- The model's ability to leverage local features and attention mechanisms enhances diagnostic capabilities.
- This method holds promise for improving the interpretation of medical imaging for thoracic diseases.
