Zeus introduces a novel framework that leverages large language models to perform medical image segmentation without requiring paired vision-language datasets.

• Enables zero-shot segmentation across different medical imaging modalities (MRI, CT)
• Integrates textual domain knowledge with visual data for improved diagnostic accuracy
• Eliminates expensive and time-consuming paired data collection requirements
• Demonstrates the power of multimodal learning in clinical applications

This breakthrough significantly advances medical imaging analysis by making sophisticated segmentation techniques more accessible and efficient, potentially accelerating diagnostic workflows in radiology and improving clinical outcomes.

Zeus: Zero-shot LLM Instruction for Union Segmentation in Multimodal Medical Imaging