Generative AI for Diagnostic Medical Imaging: A Review.

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Deep Learning

Background:

• Generative deep learning (DL) models are rapidly evolving.
• Their application in diagnostic medical imaging offers significant potential.
• Key areas include enhancing diagnostic accuracy, reducing radiation exposure, and improving data handling.

Purpose of the Study:

• To comprehensively analyze recent advancements in generative DL models for medical imaging.
• To explore various architectures like GANs, AEs, diffusion models, and transformers.
• To highlight their applications in synthetic data generation, interpretability, and image enhancement.

Main Methods:

• Review of generative adversarial networks (GANs), autoencoders (AEs), diffusion models, and transformer-based models.
• Design of pipeline architectures for medical imaging applications, including enhanced GANs (ML-C-GAN, Temporal-GAN) and AE-GAN hybrids (Atten-AE, M3AE).
• Focus on text-to-image, image-to-text translation, and image-to-image enhancement.

Main Results:

• Demonstration of clinically relevant, high-fidelity synthetic image generation across modalities.
• Successful application of models for enhancing image synthesis and diagnostic reporting.
• Showcasing capacity for accurate caption generation and patient-specific image interpretation.

Conclusions:

• Architectural innovations in generative models are crucial for medical imaging.
• These models enhance image synthesis, diagnostic reporting, and interpretation.
• Future directions aim to refine models for more accurate, accessible, and personalized patient care.