Survey on sampling conditioned brain images and imaging measures with generative models.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Generative models are powerful AI tools for creating synthetic data.
• Brain imaging data is crucial for neuroscience but faces acquisition challenges.
• Existing datasets often lack diversity, especially for rare diseases or specific demographics.

Purpose of the Study:

• To provide a comprehensive overview of generative models in brain imaging.
• To emphasize the advancements and applications of conditional generative methods.
• To discuss challenges and future directions for integrating these models into neuroscience.

Main Methods:

• Overview of deep learning-based generative models: Variational Autoencoders (VAEs), Generative Adversarial Networks (GANs), and diffusion models.
• Focus on conditional generative models that use variables like age, sex, or clinical phenotypes.
• Categorization of existing approaches and discussion of key challenges.

Main Results:

• Generative models synthesize realistic brain images with biological and clinical relevance.
• Conditional models enhance dataset diversity, enabling study of underrepresented scenarios and disease progression.
• Synthetic data addresses data privacy concerns by providing non-identifiable information.

Conclusions:

• Generative models significantly advance neuroscience by augmenting datasets and improving research capabilities.
• Conditional generative models offer solutions for data scarcity, imbalance, and privacy in brain imaging.
• These models hold potential for improving diagnostic accuracy and accelerating personalized treatment development.