Agentic AI Pipeline Enables Training-Free Brain MRI Analysis Using LLMs

A new agentic AI pipeline enables training-free automated analysis of brain MRI scans by orchestrating specialized external tools through large language models. This approach addresses the fundamental limitation of current LLM architectures, which lack native 3D spatial reasoning capabilities required for direct volumetric medical imaging analysis. The system autonomously executes complex end-to-end radiological workflows including preprocessing steps like skull stripping and registration, along with pathology segmentation for conditions such as glioma and meningioma. Researchers validated the methodology using several advanced LLMs including GPT-5.1, Gemini 3 Pro, and Claude Sonnet 4.5 combined with off-the-shelf domain-specific tools. While agentic AI offers promising solutions by eliminating the need for intrinsic 3D processing capabilities in LLMs, its feasibility in complex multi-step radiological workflows remains underexplored. The research demonstrates that state-of-the-art LLMs show high performance in general visual question answering but require alternative approaches for medical imaging applications. The work was announced as arXiv:2604.16729v1 with cross-announcement type, presenting a novel approach to neuro-radiological image analysis without requiring model training.