Researchers reviewed how artificial intelligence (AI) is being used to manage corneal endothelial diseases and related surgeries. The review looked at how AI handles tasks like measuring cell density, mapping eye thickness, and identifying issues like graft detachment. The findings show that AI is currently most reliable for these types of measurement and quality control tasks.
However, the technology faces challenges when it tries to predict future health outcomes. For example, the accuracy of AI models dropped significantly when tested on different devices or by different surgeons. Additionally, many studies lacked external validation, meaning the results were not tested in diverse, real-world settings.
Because of these inconsistencies, AI is currently more useful for technical measurements than for predicting long-term patient outcomes. Patients and doctors should view these tools as helpful for data collection and quality control rather than as definitive tools for predicting surgery success or long-term graft survival.
Common questions
What is AI currently best at in eye care?
Artificial intelligence is currently most effective for measurement and quality-control tasks. These include tasks like cell segmentation, mapping edema and thickness, and quantifying graft detachment. These technical tasks are more mature and ready for clinical use than tasks that try to predict future health outcomes.
Is AI reliable across different clinics and doctors?
The research shows that AI performance can decline when used across different devices, surgeons, and centers. Because of these variations, the technology may not perform consistently in every clinical setting. This is a key reason why it is currently used more for measurement than for making predictions.
Can AI predict long-term surgery outcomes?
The evidence suggests that AI is not yet reliable for predicting outcomes like long-term graft survival or the need for specific procedures like rebubbling. While it is useful for measuring current eye conditions, its ability to predict future complications is still limited and requires more study.
