3208 - Language-Guided Agentic AI Framework for Image Registration in Prostate SBRT Using Model Context Protocol

Purpose/Objective(s): To develop and validate an agentic AI framework using Model Context Protocol (MCP) that enables clinicians to control CT-MR image registration through natural language guidance. The study aims to determine if text-based clinical intent produces predictable, structure-specific alignment outcomes for prostate stereotactic body radiation therapy (SBRT) treatment planning.

Materials/Methods: An MCP server was developed to provide specialized registration tools for a Large Language Model agent to invoke dynamically. The framework utilized natural language parsing to identify alignment intent and target structures. For this study, four prostate SBRT patients with pre-existing prostate, bladder, and femoral head contours on both CT and MRI were used. Structure-specific algorithms included centroid matching with iterative refinement, dual-structure interface weighting, masked mutual information on bony landmarks, and multi-resolution intensity-based registration. Cases were processed under several language-driven conditions: prostate-guided, prostate-bladder interface, bony anatomy, mutual information, and directional offset adjustments.

Results: Each alignment method consistently produced distinct registration transforms, confirmed through visualization and translation matrix analysis. Structure-specific alignments improved 95th percentile Hausdorff distance (HD95) by 7-24% for their target organs compared to mutual information registration. Alignments resulted in 2-10 mm shifts from baseline translation, confirming that different language instructions produce unique registration outcomes. Directional commands produced predictable translations with corresponding changes visible in the overlay visualizations. Structure-based alignments were completed in under 10 seconds, while mutual information registration required 1-2 minutes.

Conclusion: This work demonstrates a consistent relationship between natural language guidance and registration outcomes. The framework allows clinical teams to achieve anatomical alignment goals through conversational interaction rather than manual parameter adjustment. This MCP-based architecture serves as a foundation for more complex autonomous workflows, where similar natural language interfaces could be utilized for radiotherapy treatment planning.