A Validity Analysis of Text-to-Image Generative Artificial Intelligence Models for Craniofacial Anatomy Illustration

Area of Science:

• Medical Education Technology
• Artificial Intelligence in Anatomy
• Digital Illustration

Background:

• Anatomically accurate illustrations are vital for medical education.
• Generative Artificial Intelligence (GAI) offers potential for automating illustration creation.
• This study assesses GAI for craniofacial anatomy illustration.

Purpose of the Study:

• Evaluate GAI models for generating craniofacial anatomy illustrations.
• Compare performance of different GAI models (Midjourney, DALL-E 3, Gemini, Stable Diffusion).
• Assess anatomical accuracy, aesthetics, usability, and cost-effectiveness.

Main Methods:

• Generated 736 craniofacial anatomy images using four GAI models.
• Images produced in oil painting and realistic photograph styles.
• Evaluated by four reviewers for detail, aesthetics, usability, and cost.

Main Results:

• DALL-E 3 excelled in anatomical detail and usability; Midjourney in aesthetics and cost.
• All models exhibited flaws in depicting critical anatomical features like foramina and neurovascular structures.
• High inter-rater reliability (ICC = 0.858) confirmed consistent evaluations.

Conclusions:

• GAI shows promise for rapid craniofacial anatomy illustration but has limitations.
• Inadequate training data and anatomical understanding hinder current GAI performance.
• Model refinement, expert feedback, and ethical considerations are essential for safe implementation.