Synthetic Data Generation via Generative Adversarial Networks in Healthcare: A Systematic Review of Image- and...

Area of Science:

• Artificial Intelligence
• Medical Informatics
• Machine Learning

Background:

• Generative Adversarial Networks (GANs) are advanced AI tools for unsupervised learning.
• GANs are increasingly applied in healthcare for complex data challenges.

Purpose of the Study:

• To systematically review GAN applications in healthcare, focusing on image and signal-based studies.
• To analyze utilized GAN architectures, datasets, and clinical domains.

Main Methods:

• Systematic literature review following PRISMA guidelines.
• Analysis of 72 relevant journal articles on GANs in healthcare.
• Categorization of studies by clinical domain, data type, and GAN architecture.

Main Results:

• Magnetic Resonance Imaging (MRI) and Electrocardiogram (ECG) data were most frequently studied.
• Brain, cardiology, cancer, ophthalmology, and lung studies dominated research areas.
• Conditional GAN (cGAN) and CycleGAN were the most common architectures.
• Promising results were observed in data augmentation, anonymization, and multi-task learning.

Conclusions:

• GANs offer significant potential for advancing healthcare through improved data handling and analysis.
• Standardization of evaluation metrics and direct performance comparisons are needed.
• Future research should focus on no-reference metrics, simulation, and interpretability.