Comparison of anatomy image generation capability in AI image generation models.

Abstract

The advancement of artificial intelligence (AI) has significantly impacted various fields, and in recent years, high-performing AI image generation models have emerged. This paper explores the capabilities of these models, specifically DALL-E 2, Midjourney 5, and Stable Diffusion 1.5, in generating anatomical images where accurate depiction is crucial rather than mere creativity. The study evaluates the learning extent of anatomical terminology and the anatomical accuracy of generated images by these models across three main categories: bones, organs, and muscles. Additionally, a comparison was made a year later using the advanced versions of two models, Midjourney 6 and DALL-E 3, which had been reported to show significant improvements in image quality over their previous versions. However, even with these improvements, we conclude that AI models cannot fully replace the expertise, communication skills, and creative judgement of professional medical illustrators. This study emphasises that using AI as a complementary tool can enhance the quality of anatomical and medical communications and education, and this approach helps predict the future impact on traditional medical illustration fields.