Controllable Medical Image Generation via GAN.

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Authentic medical image data is scarce, limiting research in areas like medical image perception and clinician training.
• Existing datasets are often difficult to use due to extensive processing requirements (de-identification, labeling).
• Previous artificial medical images lacked authenticity, raising concerns about clinical applicability.

Purpose of the Study:

• To propose a novel method for generating authentic medical images using Generative Adversarial Networks (GANs).
• To enable controllable manipulation of generated medical image attributes for diverse research needs.
• To address the scarcity and processing challenges of authentic medical imaging data for research applications.

Main Methods:

• Utilized Generative Adversarial Networks (GANs) to synthesize high-quality, authentic medical images.
• Implemented a controllable manipulation technique to adjust image attributes according to experimental requirements.
• Validated the method across various medical imaging modalities, including mammograms, MRI, CT, and skin cancer images.

Main Results:

• Successfully generated authentic medical images that mimic real-world data.
• Demonstrated the controllability of image attributes for tailored experimental designs.
• Verified the effectiveness of the GAN-based approach for diverse medical imaging modalities.

Conclusions:

• The proposed GAN-based method effectively generates authentic and controllable medical images.
• This approach can significantly benefit medical image perception research by providing readily available, realistic datasets.
• The generated images and model are suitable for various medical imaging studies, enhancing research generalizability and efficiency.