Scholarlink

Class imbalance and a lack of diverse data are significant challenges in medical image analysis. To tackle these issues, we introduce a generative model that combines a Vision Transformer with a Denoising Diffusion Probabilistic Model to produce high-quality medical images. We evaluated the model using the BreaKHis (breast histopathology) and HAM10000 (skin lesion) datasets. The synthetic images created by the ViT-DDPM were used to expand the training data for a ResNet-152 classifier. The results show that the generated images enhance classification performance and help reduce class imbalance. This work emphasizes the potential of merging diffusion models with transformer architectures to support fair and effective diagnostic systems in medical imaging.

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

High-Quality Medical Image Synthesis and Class Imbalance Handling with Denoising Diffusion Probabilistic Models

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Anubhab Das,
  • Sweta Jain,
  • B. N. Roy

摘要

Class imbalance and a lack of diverse data are significant challenges in medical image analysis. To tackle these issues, we introduce a generative model that combines a Vision Transformer with a Denoising Diffusion Probabilistic Model to produce high-quality medical images. We evaluated the model using the BreaKHis (breast histopathology) and HAM10000 (skin lesion) datasets. The synthetic images created by the ViT-DDPM were used to expand the training data for a ResNet-152 classifier. The results show that the generated images enhance classification performance and help reduce class imbalance. This work emphasizes the potential of merging diffusion models with transformer architectures to support fair and effective diagnostic systems in medical imaging.