Related Experiment Video
Updated: Jun 12, 2025
07:35
Corneal Tissue Engineering: An In Vitro Model of the Stromal-nerve Interactions of the Human Cornea
Published on: January 24, 2018
8.8K
Assessment of Generative Artificial Intelligence (AI) Models in Creating Medical Illustrations for Various Corneal
Kayvon A Moin1,2, Ayesha A Nasir3, Dallas J Petroff4
1Hoopes Vision Research Center, Hoopes Vision, Draper, USA.
Cureus
|September 27, 2024
Summary
Generative AI models like DALL-E 3 and MIM struggle to create accurate medical illustrations for corneal transplant procedures, scoring significantly lower than human-created images. Further AI development is needed for reliable ophthalmology visuals.
Area of Science:
- Ophthalmology
- Medical Illustration
- Artificial Intelligence
Background:
- Generative AI is increasingly used for various tasks, including image creation.
- Accurate medical illustrations are crucial for surgical training and patient education in ophthalmology.
Purpose of the Study:
- To assess the capability of generative AI models (DALL-E 3, MIM) in creating medical illustrations for corneal transplant procedures.
- To compare AI-generated illustrations against human-created controls using a standardized grading system.
Main Methods:
- Six AI models were prompted to generate illustrations for Descemet's stripping automated endothelial keratoplasty (DSAEK), Descemet's membrane endothelial keratoplasty (DMEK), deep anterior lamellar keratoplasty (DALK), and penetrating keratoplasty (PKP).
- Illustrations were evaluated by independent reviewers based on legibility, anatomical accuracy, procedural accuracy, and fictitious anatomy.
- AI model ChatGPT-4o was used to grade the generated and control illustrations.
Main Results:
- Control illustrations significantly outperformed AI-generated images from DALL-E 3 and MIM in overall quality and accuracy (p<0.001).
- AI models achieved significantly lower mean cumulative scores (DALL-E 3: 29.2%, MIM: 37.5%) compared to control illustrations (97.1%).
- ChatGPT-4o overestimated the quality of AI-generated images, while accurately assessing human-created illustrations.
Conclusions:
- Current generative AI models are limited in producing medically accurate illustrations for corneal transplant procedures.
- Significant advancements in AI technology are necessary to ensure the reliability and accuracy of medical visuals in ophthalmology.
