Scholarlink

<p>Deep learning (DL) has transformed medical image analysis by automating complex feature extraction and improving accuracy across diverse diagnostic tasks. Recent advances in neural architectures such as Convolutional Neural Networks (CNNs), U-Net, ResNet, DenseNet, Generative Adversarial Networks (GANs), and Vision Transformers (ViTs) have enabled high-performance segmentation, classification, detection, and reconstruction in medical imaging. This survey reviews the key developments categorizing models by application and discussing benchmark datasets that have standardized evaluation in the field. It further highlights emerging directions including self-supervised and multimodal learning, federated privacy-preserving frameworks, and diffusion-based generative methods that address data scarcity and domain generalization. Remaining challenges such as interpretability, bias, and regulatory compliance are also examined. The paper concludes with future perspectives on building clinically explainable, ethically governed, and robust DL systems capable of transforming healthcare diagnostics and precision medicine.</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

Recent Advances in Medical Image Analysis Using Deep Learning: A Survey

  • 期刊论文

  • 发表日期：

    2026年4月17日

查看全文
  • Bhawna Goyal,
  • Ayush Dogra,
  • Dawa Chyophel Lepcha,
  • Ahmed Alkhayyat,
  • Murari Devakannan Kamalesh,
  • Sarbeswara Hota,
  • Ritu Sharma

摘要

Deep learning (DL) has transformed medical image analysis by automating complex feature extraction and improving accuracy across diverse diagnostic tasks. Recent advances in neural architectures such as Convolutional Neural Networks (CNNs), U-Net, ResNet, DenseNet, Generative Adversarial Networks (GANs), and Vision Transformers (ViTs) have enabled high-performance segmentation, classification, detection, and reconstruction in medical imaging. This survey reviews the key developments categorizing models by application and discussing benchmark datasets that have standardized evaluation in the field. It further highlights emerging directions including self-supervised and multimodal learning, federated privacy-preserving frameworks, and diffusion-based generative methods that address data scarcity and domain generalization. Remaining challenges such as interpretability, bias, and regulatory compliance are also examined. The paper concludes with future perspectives on building clinically explainable, ethically governed, and robust DL systems capable of transforming healthcare diagnostics and precision medicine.