[Research progress on medical image dataset expansion methods].

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer-Aided Diagnosis

Background:

• Computer-aided diagnosis (CAD) systems require diverse training data for optimal performance.
• Current medical image datasets are often limited by imaging costs, labeling expenses, and patient privacy concerns, leading to insufficient data diversity and acquisition challenges.
• Efficient and cost-effective medical image dataset expansion is a critical research area.

Approach:

• This review systematically analyzes existing literature on medical image dataset expansion techniques.
• It compares and contrasts methods based on geometric transformations with those utilizing generative adversarial networks (GANs).
• Emphasis is placed on advancements and improvements within GAN-based augmentation strategies.

Key Points:

• Geometric transformation methods offer basic augmentation but may lack diversity.
• Generative adversarial networks (GANs) show significant promise for creating realistic and diverse synthetic medical images.
• Improvements in GAN architectures and training methodologies are crucial for advancing medical image augmentation.

Conclusions:

• Addressing data scarcity through effective augmentation is vital for enhancing CAD system accuracy and reliability.
• Generative adversarial networks represent a key frontier in medical image dataset expansion.
• Future research should focus on overcoming current challenges and exploring novel GAN applications in this domain.