Scholarlink

The generation of accurate medical images is necessary for the enhancement of the reliability of diagnostics and treatment protocols. This paper introduces DD-GAN, a hybrid model that unites Diffusion Models with Generative Adversarial Networks to improve image quality and reduce sampling time. Evaluation of the model is done using the BreakHis dataset, which includes Histopathological Breast cancer images labeled as Benign and Malignant tumors at various magnifications. DD-GAN enhances image quality through the use of iterative denoising techniques, employing diffusion models to address complex data distributions and utilizing GANs for their computational efficiency to balance realism and resource usage. Quantitative tests using FID, PSNR, and SSIM measures validate its capacity to produce diagnostically relevant images on par with the original data. This model facilitates scalable medical image synthesis and can be applied to other datasets with potential for diversity and accelerating the generation process.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Hybrid Diffusion Models for Scalable and High-Quality Medical Image Generation: A Computationally Efficient Approach

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Amisha Rao,
  • Sweta Jain,
  • B. N. Roy

摘要

The generation of accurate medical images is necessary for the enhancement of the reliability of diagnostics and treatment protocols. This paper introduces DD-GAN, a hybrid model that unites Diffusion Models with Generative Adversarial Networks to improve image quality and reduce sampling time. Evaluation of the model is done using the BreakHis dataset, which includes Histopathological Breast cancer images labeled as Benign and Malignant tumors at various magnifications. DD-GAN enhances image quality through the use of iterative denoising techniques, employing diffusion models to address complex data distributions and utilizing GANs for their computational efficiency to balance realism and resource usage. Quantitative tests using FID, PSNR, and SSIM measures validate its capacity to produce diagnostically relevant images on par with the original data. This model facilitates scalable medical image synthesis and can be applied to other datasets with potential for diversity and accelerating the generation process.