Related Experiment Videos
Evaluation of Generative Artificial Intelligence Models in Producing Anatomically Accurate Illustrations: A
Aleksandra Strugała1, Bartosz Batkowski1, Kamila Skrzypoń1
1Faculty of Medicine and Health Sciences, University of Applied Sciences, Tarnów, Poland.
Clinical Anatomy (New York, N.Y.)
|July 21, 2026
Summary
Generative artificial intelligence (GAI) models show limited accuracy in creating anatomically correct human illustrations. Expert review is crucial before using these AI-generated images in medical education or research.
Area of Science:
- Biomedical Sciences
- Medical Education
- Artificial Intelligence
Background:
- Generative artificial intelligence (GAI) is increasingly used in biomedical fields.
- Text-to-image generators offer potential for rapid creation of anatomical visuals.
- The anatomical accuracy of GAI-generated images is currently uncertain.
Purpose of the Study:
- To evaluate the anatomical accuracy of selected Generative artificial intelligence models.
- To assess the performance of six text-to-image generators in creating correct human anatomical illustrations.
Main Methods:
- Six Generative artificial intelligence models were tested using standardized prompts for anatomical structures.
- Images were analyzed for presence and physiological correctness by four independent reviewers against Terminologia Anatomica.
- Statistical analysis included proportion calculations, inter-rater agreement (Cohen's κ), and logistic regression.
Main Results:
- DALL-E showed the highest proportion of present (56.7%) and physiologically correct (66.0%) structures.
- Overall, fully correct anatomical illustrations were limited across all models, with DALL-E achieving 37.4%.
- Accuracy varied significantly by anatomical region, with complex structures like the lumbar plexus showing particularly low correctness.
Conclusions:
- Current Generative artificial intelligence models have substantial limitations in producing anatomically accurate medical illustrations.
- While some models perform better, the reliability of GAI-generated images is insufficient for unsupervised use.
- Expert verification is essential before applying Generative artificial intelligence-generated anatomical images in medical education or scientific contexts.
