Large Language Model-Based Interactive Code Generation for Developing a 3D Eye Movement Schematic

Area of Science:

• Ophthalmology
• Medical Education
• Computational Anatomy

Background:

• Accurate 3D geometric understanding of extraocular muscles and the globe is crucial for strabismus management.
• Static educational tools and complex biomechanical software limit clinical application.
• Current AI image generation often results in anatomical inaccuracies (hallucinations).

Purpose of the Study:

• To develop a structurally coherent, interactive 3D eye movement schematic using a large language model (LLM).
• To demonstrate the feasibility of generating 3D medical schematics via natural language dialogue.
• To create an accessible educational tool for strabismus management.

Main Methods:

• Utilized an LLM to generate web-based 3D schematic code (HTML, JavaScript/Three.js) through natural language prompts.
• Incorporated anatomical parameters from literature and mathematical constraints (quaternions, spherical linear interpolation) to ensure structural accuracy.
• Employed an iterative process of prompt refinement and debugging, validated by ophthalmologists.

Main Results:

• Successfully developed a functional, interactive 3D eye movement schematic.
• Achieved an acceptable schematic with an average of 7.4 prompt inputs per session.
• The final schematic accurately depicted ocular structures and enabled kinematic visualizations without failure, despite initial AI hallucination challenges.

Conclusions:

• LLMs can translate anatomical descriptions into spatial logic for creating structurally sound 3D medical schematics.
• This approach democratizes the creation of interactive educational tools for healthcare providers without coding expertise.
• Enables intuitive generation of customizable 3D materials for patient education and medical training.