Generative adversarial network in medical imaging: A review.

Area of Science:

• Computer Vision
• Medical Imaging
• Machine Learning

Background:

• Generative adversarial networks (GANs) are powerful for data generation without explicit probability density modeling.
• Adversarial loss in GANs enables effective use of unlabeled data and higher-order consistency.
• GANs have found broad applications in domain adaptation, data augmentation, and image-to-image translation.

Purpose of the Study:

• To review recent advances in medical imaging utilizing adversarial training schemes.
• To provide a comprehensive overview for researchers interested in GANs for medical applications.

Main Methods:

• Review of recent literature on generative adversarial networks in medical imaging.
• Categorization of applications based on adversarial training schemes.

Main Results:

• GANs are rapidly adopted in medical imaging for reconstruction, segmentation, detection, classification, and cross-modality synthesis.
• Adversarial training offers significant advantages in various medical imaging tasks.
• The trend of GAN adoption in medical imaging is expected to continue.

Conclusions:

• Generative adversarial networks are increasingly influential in medical imaging research.
• This review highlights the growing utility and diverse applications of GANs in the field.
• Future research is expected to further leverage adversarial training for medical imaging innovations.