Generative adversarial networks and its applications in the biomedical image segmentation: a comprehensive survey

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Computer Vision

Background:

• Medical image segmentation is a critical yet challenging task in biomedical imaging.
• Deep generative models, particularly Generative Adversarial Networks (GANs), show significant promise for image analysis tasks.
• Existing literature presents various GAN-based approaches for medical image segmentation.

Purpose of the Study:

• To conduct a comprehensive survey of Generative Adversarial Networks (GANs) applied to medical image segmentation.
• To analyze GAN models, performance metrics, loss functions, datasets, and augmentation techniques used in medical image segmentation.
• To provide an overview of GAN applications in segmenting images related to various human diseases.

Main Methods:

• Systematic literature search identifying 151 relevant papers.
• Two-stage screening process to select 138 papers for final analysis.
• Qualitative synthesis of selected papers focusing on GAN architectures, implementation details, and disease-specific applications.

Main Results:

• Identified and categorized various GAN-based models used for medical image segmentation.
• Detailed the common performance metrics, loss functions, and datasets employed in GAN-based segmentation.
• Highlighted the application of GANs across diverse medical imaging modalities and disease types.
• Reviewed implementation aspects including source code availability.

Conclusions:

• GANs offer substantial potential for advancing medical image segmentation accuracy and efficiency.
• The survey critically discusses current limitations of GANs in this domain.
• Identified key areas and provided suggestions for future research directions in GANs for biomedical image segmentation.