Scholarlink

<p>The emergence of deep learning has propelled artificial intelligence (AI) into a truly new stage of development. With the rapid iteration of this technology, massive datasets have been continuously proposed. Medical artificial intelligence (MAI) has evolved from traditional machine learning to deep learning. Vision<b>–</b>language foundation models (VLM), the core of multimodal intelligence, excel in cross-modal fusion: VLM can simultaneously integrate and process information from visual modalities (e.g., images, videos, and audio) and textual modalities, enabling high-level collaborative visual and linguistic tasks. Compared with unimodal systems, the performance of multimodal architectures has achieved significant improvement. Leveraging visual computing, VLM realize precise cross-modal alignment, highly matching healthcare’s demand for multi-source visual-text information integration and supporting clinical applications like real-time diagnosis. However, current healthcare VLM research lacks a systematic visual computing perspective review. This paper explores its challenges and future from visual computing evolution, constructing a full-link technical roadmap, analyzing unique bottlenecks and proposing a 3D development framework. It also outlines relevant research elements, analyzes multi-dimensional challenges and prospects VLM-driven agentic AI systems, laying a foundation for healthcare VLM application and medical visual computing upgrading.</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

Vision–language foundation model driven agentic AI systems for healthcare

  • 期刊论文

  • 发表日期：

    2026年3月26日

查看全文
  • Lifeng Chen,
  • Xinming Xu,
  • Yiming Qin,
  • Haoxuan Li,
  • Nan Jiang

摘要

The emergence of deep learning has propelled artificial intelligence (AI) into a truly new stage of development. With the rapid iteration of this technology, massive datasets have been continuously proposed. Medical artificial intelligence (MAI) has evolved from traditional machine learning to deep learning. Vision–language foundation models (VLM), the core of multimodal intelligence, excel in cross-modal fusion: VLM can simultaneously integrate and process information from visual modalities (e.g., images, videos, and audio) and textual modalities, enabling high-level collaborative visual and linguistic tasks. Compared with unimodal systems, the performance of multimodal architectures has achieved significant improvement. Leveraging visual computing, VLM realize precise cross-modal alignment, highly matching healthcare’s demand for multi-source visual-text information integration and supporting clinical applications like real-time diagnosis. However, current healthcare VLM research lacks a systematic visual computing perspective review. This paper explores its challenges and future from visual computing evolution, constructing a full-link technical roadmap, analyzing unique bottlenecks and proposing a 3D development framework. It also outlines relevant research elements, analyzes multi-dimensional challenges and prospects VLM-driven agentic AI systems, laying a foundation for healthcare VLM application and medical visual computing upgrading.