Scholarlink

Deep learning in medical imaging often requires large-scale, high-quality data, or initiation with suitably pre-trained weights. However, medical datasets are limited by data availability, domain-specific knowledge, and privacy concerns, and the creation of large and diverse radiologic databases like RadImageNet is highly resource-intensive. To address these limitations, we introduce RadImageGAN, a multi-modal medical image generator developed by training StyleGAN-XL on the RadImageNet dataset of CT and MRI images and the HyperKvasir dataset of gastrointestinal images. RadImageGAN can generate high-resolution synthetic medical imaging datasets across 12 anatomical regions and 130 pathological classes in 3 modalities. Furthermore, we developed the RadImageGAN-Labeler, which can generate multi-class pixel-wise annotated paired synthetic images and masks for diverse downstream segmentation tasks with minimal manual annotation. Using synthetic auto-labeled data from RadImageGAN can significantly improve performance on four diverse downstream segmentation datasets by augmenting real training data. We find that RadImageGAN can improve model performance and address data scarcity while reducing the resources needed for annotations for segmentation tasks.

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

RadImageGAN – A Multi-modal Dataset-Scale Generative AI for Medical Imaging

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Zelong Liu,
  • Peyton Smith,
  • Alexander Lautin,
  • Jieshen Zhou,
  • Maxwell Yoo,
  • Mikey Sullivan,
  • Haorun Li,
  • Louisa Deyer,
  • Alexander Zhou,
  • Arnold Yang,
  • Alara Yimaz,
  • Catherine Zhang,
  • James Grant,
  • Daiqing Li,
  • Zahi A. Fayad,
  • Sean Huver,
  • Timothy Deyer,
  • Xueyan Mei

摘要

Deep learning in medical imaging often requires large-scale, high-quality data, or initiation with suitably pre-trained weights. However, medical datasets are limited by data availability, domain-specific knowledge, and privacy concerns, and the creation of large and diverse radiologic databases like RadImageNet is highly resource-intensive. To address these limitations, we introduce RadImageGAN, a multi-modal medical image generator developed by training StyleGAN-XL on the RadImageNet dataset of CT and MRI images and the HyperKvasir dataset of gastrointestinal images. RadImageGAN can generate high-resolution synthetic medical imaging datasets across 12 anatomical regions and 130 pathological classes in 3 modalities. Furthermore, we developed the RadImageGAN-Labeler, which can generate multi-class pixel-wise annotated paired synthetic images and masks for diverse downstream segmentation tasks with minimal manual annotation. Using synthetic auto-labeled data from RadImageGAN can significantly improve performance on four diverse downstream segmentation datasets by augmenting real training data. We find that RadImageGAN can improve model performance and address data scarcity while reducing the resources needed for annotations for segmentation tasks.