Generative Adversarial Networks in Medical Image augmentation: A review

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Computer Vision

Background:

• Deep learning models for medical diagnosis require extensive training data.
• Generative Adversarial Networks (GANs) show promise for medical image generation and data augmentation.
• Limited training samples pose a significant challenge in medical image-based AI development.

Purpose of the Study:

• To systematically review and analyze the current state of medical image augmentation research.
• To provide a comprehensive overview of Generative Adversarial Network (GAN)-based augmentation techniques in the medical field.
• To identify research trends, limitations, and future prospects in medical image augmentation.

Main Methods:

• A systematic literature review of 105 papers on medical image augmentation from 2018-2021 was conducted.
• Papers were analyzed based on organ-specific images, datasets, loss functions, and evaluation metrics.
• Key journals and conferences in medical image processing were also surveyed.

Main Results:

• Various augmentation models, loss functions, and evaluation metrics were summarized, offering a reference for researchers.
• The relationship between augmented models and training set size was explored, particularly for limited-quality datasets.
• Statistical analysis indicates strong development momentum in this research field.

Conclusions:

• GAN-based medical image augmentation effectively addresses the scarcity of training samples for diagnostic and treatment models.
• This review provides valuable insights for researchers focusing on GAN applications in medical imaging.
• Further research directions and existing model limitations are discussed to guide future work.