Assessing the Role of Artificial Intelligence in the Creation of Patient Educational Videos for Corneal Refractive Surgery.

PURPOSE

We aim to assess the ability of artificial intelligence (AI) to generate patient educational videos for various corneal refractive surgeries.

METHODS

Three AI text-to-video platforms (InVideo (San Francisco, CA), ClipTalk (San Francisco, CA), and EasyVid (Los Angeles, CA)) were used to create patient educational videos for laser-assisted in situ keratomileusis (LASIK), photorefractive keratectomy (PRK), and small incision lenticule extraction (SMILE), respectively. Videos for LASIK and PRK from the American Academy of Ophthalmology (AAO) and a SMILE video from Zeiss served as controls for each surgery. A three-point grading system (from zero to three, with zero being the worst and three being the best in each category) was used to compare videos in terms of "image accuracy," "script accuracy," "image clarity," and "script alignment."

RESULTS

In terms of image accuracy, the control videos outperformed InVideo, EasyVid, and ClipTalk for LASIK (3 versus 0.667 versus 0 versus 0; p<0.005), PRK (3 versus 1 versus 0.33 versus 0; p<0.05 for InVideo, p<0.005 all), and SMILE (3 versus 0.33 versus 0 versus 0.33; p<0.005), respectively. With a few exceptions, all three AI models performed similarly to the control videos in terms of script accuracy, image clarity, and script alignment.

CONCLUSION

In their current state, AI text-to-video generators can produce surgical educational videos for patients with accurate script narration and high image clarity, although these platforms are not yet capable of producing medically accurate images to go along with these scripts. Further improvements in the medical accuracy of these images must be made to make these videos more appropriate for patient consumption.