Advanced image generation for cancer using diffusion models.

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Oncology

Background:

• Deep neural networks (DNNs) show promise in medical image analysis but are limited by small datasets.
• Generative models, especially diffusion models, can synthesize photorealistic images, expanding AI applications in medicine.

Purpose of the Study:

• To investigate the use of diffusion models for generating high-quality medical images, including brain MRI, mammography, and chest X-rays.
• To assess the ability of diffusion models to capture oncology-specific attributes and preserve patient anonymity.

Main Methods:

• Utilized the DreamBooth platform to train stable diffusion models.
• Employed text prompts, class images, and instance images for model training.
• Evaluated synthesized image quality using the Fréchet inception distance (FID) metric.

Main Results:

• Successfully generated diverse, high-quality medical images across multiple modalities (brain MRI, CESM, chest/lung X-ray).
• Demonstrated high fidelity between synthesized and real images via FID scores.
• Confirmed the preservation of oncology-specific attributes in generated images.

Conclusions:

• Diffusion models offer a robust framework for generating oncological medical imagery, addressing data limitations in AI.
• This AI-driven approach enhances patient anonymity and mitigates re-identification risks in data sharing.
• The study establishes a foundation for integrating advanced generative AI in medical imaging research and development.