Scholarlink

<p>Generative Artificial Intelligence (AI) has emerged as a transformative paradigm in medical imaging, enabling advanced capabilities for image synthesis, reconstruction, and enhancement across radiological modalities. This review provides a structured, modality-wise analysis of recent developments in generative AI, focusing on applications in X-ray, computed tomography (CT), magnetic resonance imaging (MRI), positron emission tomography (PET), and ultrasound. Key models, including Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), diffusion models, and transformer-based architectures, are evaluated for tasks such as super-resolution, modality translation, artifact removal, and data augmentation. A systematic literature review was conducted across IEEE Xplore, PubMed, Scopus, and ScienceDirect for studies published between 2021 and 2025. Using relevant keywords, approximately 135 studies were identified, of which 105 were screened, and 32 were selected for detailed analysis based on predefined inclusion criteria. The findings indicate that generative AI significantly improves image quality and diagnostic performance while addressing challenges such as data scarcity and radiation dose reduction. However, limitations including dataset shift, hallucination risk, and lack of multi-centre validation restrict clinical deployment. Future research should focus on standardized evaluation, large-scale multi-institutional validation, federated learning, and physics-informed models to enhance robustness and clinical reliability.</p>

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

Generative AI for Medical Imaging: A Systematic Review of Models, Applications, and Clinical Challenges

  • 期刊论文

  • 发表日期：

    2026年6月23日

查看全文
  • Apurva A. Khandekar,
  • Deepti D. Shrimankar,
  • Reetu Gupta

摘要

Generative Artificial Intelligence (AI) has emerged as a transformative paradigm in medical imaging, enabling advanced capabilities for image synthesis, reconstruction, and enhancement across radiological modalities. This review provides a structured, modality-wise analysis of recent developments in generative AI, focusing on applications in X-ray, computed tomography (CT), magnetic resonance imaging (MRI), positron emission tomography (PET), and ultrasound. Key models, including Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), diffusion models, and transformer-based architectures, are evaluated for tasks such as super-resolution, modality translation, artifact removal, and data augmentation. A systematic literature review was conducted across IEEE Xplore, PubMed, Scopus, and ScienceDirect for studies published between 2021 and 2025. Using relevant keywords, approximately 135 studies were identified, of which 105 were screened, and 32 were selected for detailed analysis based on predefined inclusion criteria. The findings indicate that generative AI significantly improves image quality and diagnostic performance while addressing challenges such as data scarcity and radiation dose reduction. However, limitations including dataset shift, hallucination risk, and lack of multi-centre validation restrict clinical deployment. Future research should focus on standardized evaluation, large-scale multi-institutional validation, federated learning, and physics-informed models to enhance robustness and clinical reliability.