Related Experiment Video
Updated: May 24, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.6K
Enhancing generalization of medical image segmentation via game theory-based domain selection
Zuyu Zhang1, Yan Li2, Byeong-Seok Shin2
1Key Laboratory of Big Data Intelligent Computing, Chongqing University of Posts and Telecommunications, Chongqing, 400065, China.
Journal of Biomedical Informatics
|March 6, 2025
Summary
This study introduces a novel game theory approach for medical image segmentation, enhancing model generalization across diverse datasets. The method improves robustness against domain shifts, leading to better performance in tasks like polyp and prostate segmentation.
Area of Science:
- Medical Imaging
- Computer Vision
- Artificial Intelligence
Background:
- Medical image segmentation models struggle with generalization due to variations in imaging, anatomy, and demographics.
- Existing domain generalization methods often neglect performance balance across domains, limiting effectiveness.
- Suboptimal generalization hinders the reliable application of AI in diverse clinical settings.
Purpose of the Study:
- To develop a novel approach for improving the generalization of medical image segmentation models.
- To enhance model adaptability and robustness against domain shifts using game theory.
- To achieve balanced performance across different medical imaging datasets.
Main Methods:
- Modeling the training process as a zero-sum game to reach a Nash equilibrium.
- Implementing an adaptive domain selection method guided by the Beta distribution.
- Optimizing the adaptive selection using reinforcement learning for dynamic adjustment.
Main Results:
- The proposed method demonstrated improved generalization performance on benchmark datasets.
- Achieved an average Dice score improvement of 1.75% compared to existing methods.
- Successfully applied to polyp, optic cup/optic disc, and prostate segmentation tasks.
Conclusions:
- The game theory-based approach significantly enhances the generalization capabilities of medical image segmentation models.
- Adaptive domain selection is crucial for maintaining performance balance and robustness.
- This method offers a promising direction for developing more reliable AI tools in medical imaging.
