Back to all papers

A Multi-Agent MLLM Framework for Imaging-Grounded Treatment Recommendation in Acute Ischemic Stroke.

July 29, 2026pubmed logopapers

Authors

Yan B,Li Z,Fan Y,Li Y,Chen L,Song X,Tang Y,Shen L,Zhang R,Li Y

Affiliations (10)

  • Department of Diagnostic and Interventional Radiology, Shanghai Sixth People's Hospital Affiliated to Shanghai Jiao Tong University School of Medicine, Shanghai, China.
  • Faculty of Medical Imaging Technology, College of Health Science and Technology, Shanghai Jiao Tong University School of Medicine, Shanghai, China.
  • School of Information and Intelligent Science, Donghua University, Shanghai, China.
  • Shenyang University of Technology, Shenyang, China.
  • Department of Radiology, National Children's Medical Center, Shanghai Jiao Tong University School of Medicine, Shanghai, China.
  • Department of Radiology, Jinshan Hospital of Affiliated to Fudan University, Shanghai, China.
  • Department of Neurology, Shanghai Sixth People's Hospital Affiliated to Shanghai Jiao Tong University School of Medicine, Shanghai, China.
  • Clinical Research Center, Shanghai Sixth People's Hospital Affiliated to Shanghai Jiao Tong University School of Medicine, Shanghai, China.
  • Department of Diagnostic and Interventional Radiology, Shanghai Sixth People's Hospital Affiliated to Shanghai Jiao Tong University School of Medicine, Shanghai, China. [email protected].
  • Department of Diagnostic and Interventional Radiology, Shanghai Sixth People's Hospital Affiliated to Shanghai Jiao Tong University School of Medicine, Shanghai, China. [email protected].

Abstract

Acute ischemic stroke (AIS) requires rapid integration of multimodal imaging, contraindication screening, and treatment selection within minutes, yet real-world decision-making is limited by fragmented workflows and uneven specialist availability. We developed an imaging-grounded multi-agent multimodal large language model (MLLM) framework that emulates a multidisciplinary stroke team for structured treatment recommendation. The framework decomposes the AIS workflow into coordinated agents for triage, time-window assessment, multimodal imaging interpretation, contraindication screening, and final treatment recommendation, with retrieval-augmented generation (RAG) incorporating evidence-grounded knowledge. We retrospectively evaluated the framework in 1,658 consecutive suspected-AIS cases from two centers. Compared with a single-prompt baseline, treatment recommendation accuracy improved from 0.636 to 0.755 with multi-agent coordination and further to 0.785 after RAG augmentation, while Macro-F1 increased from 0.562 to 0.732 and 0.768, respectively. The framework showed improvements in reperfusion pathways, more effectively retaining candidates for intravenous thrombolysis and mechanical thrombectomy. Imaging agent yielded sensitivity/specificity of 0.490/0.999 for non-contrast CT intracranial hemorrhage detection, 0.883/0.851 for CT angiography large-vessel occlusion identification, and 0.939/0.739 for CT perfusion deficit assessment. These results suggest that structured imaging-grounded multi-agent coordination enhances reliability and clinical alignment of MLLM-based decision support in AIS, supporting potential deployment in time-critical stroke care settings with limited specialist availability.

Topics

Journal Article

Ready to Sharpen Your Edge?

Subscribe to join 11k+ peers who rely on RadAISlice. Get the essential weekly briefing that empowers you to navigate the future of radiology.

We respect your privacy. Unsubscribe at any time.