Related Experiment Video
Updated: Jun 30, 2026
02:09
Multi-modal Pulmonary Imaging: Using Complementary Information from CT and Hyperpolarized 129Xe MRI to Evaluate Lung Structure-Function
Published on: April 12, 2024
A Two-Stage Contrastive Learning Framework Grounded in Label-Specific Features for Low-Frequency Labels in Chest
Shi Tang1, Meiyan Huang1, Qianjin Feng1,2,3
1School of Biomedical Engineering, Southern Medical University, Guangzhou 510515, China.
Bioengineering (Basel, Switzerland)
|May 27, 2026
Summary
This study introduces a novel dual-phase convolutional neural network for improved thoracic disease classification from chest X-rays. The model effectively addresses data imbalances, enhancing diagnostic accuracy for critical respiratory conditions.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer Vision
Background:
- Thoracic diseases pose a significant health risk.
- Chest X-ray imaging is a cost-effective primary diagnostic tool.
- Existing classification models struggle with imbalanced disease label distributions.
Purpose of the Study:
- To develop an advanced dual-phase convolutional neural network for thoracic disease classification.
- To overcome limitations of current models in handling imbalanced datasets.
- To improve the accuracy and generalization of chest X-ray diagnostic tools.
Main Methods:
- A dual-phase convolutional neural network architecture was proposed.
- Phase one utilized matrix operations for label-specific feature extraction.
- Phase two incorporated feature contrastive loss and updating mechanisms for enhanced generalization.
Main Results:
- The model achieved an AUC of 0.8296, AUPRC of 0.2969, Precision of 0.3943, and F1-score of 0.3301.
- Performance was validated across three public datasets (CheXpert, REFLACX, EGD).
- The proposed model outperformed existing chest X-ray classification methods.
Conclusions:
- The framework effectively learns label-specific characteristics and intrinsic image features.
- The dual-phase network offers an advanced technical solution for thoracic disease diagnosis.
- This approach enhances the reliability of chest X-ray analysis in clinical settings.
