Related Experiment Video
Updated: May 6, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.7K
Dynamic Weighting Translation Transfer Learning for Imbalanced Medical Image Classification.
Chenglin Yu1,2, Hailong Pei3
1School of Electrtronic & Information Engineering and Communication Engineering, Guangzhou City University of Technology, Guangzhou 510800, China.
Entropy (Basel, Switzerland)
|May 24, 2024
Summary
This study introduces Dynamic Weighting Translation Transfer Learning (DTTL) to improve imbalanced medical image classification. DTTL effectively addresses domain shift and class imbalance, enhancing diagnostic model performance in clinical settings.
Area of Science:
- Artificial Intelligence
- Medical Imaging
- Machine Learning
Background:
- Deep learning shows promise in medical image diagnosis.
- Real-world applications face challenges like domain shift and class imbalance.
- These issues lead to biased models and invalidated performance on new datasets.
Purpose of the Study:
- To propose a novel transfer learning solution for imbalanced medical image classification.
- To address domain shift and class imbalance in medical image diagnosis.
- To enhance the practical applicability of deep learning models in clinical medicine.
Main Methods:
- Dynamic Weighting Translation Transfer Learning (DTTL) framework.
- Modules include Cross-domain Discriminability Adaptation (CDA), Dynamic Domain Translation (DDT), and Balanced Target Learning (BTL).
- Utilizes information and entropy theory, synthetic data generation, and confidence-based selection.
Main Results:
- DTTL significantly enhances imbalanced medical image classification performance.
- The method effectively mitigates domain shift and class imbalance issues.
- Achieved superior performance compared to existing state-of-the-art methods in extensive experiments.
Conclusions:
- DTTL offers a robust solution for practical medical image diagnosis.
- Innovates the application of entropy and information theory in deep learning for medical imaging.
- Demonstrates improved diagnostic model accuracy and reliability in challenging clinical scenarios.
