Researchers looked at 10 studies involving nearly 11,000 eyes to see if artificial intelligence could predict which people with keratoconus (a condition where the cornea gets cone-shaped) would get worse over time. Keratoconus is an eye disease that can blur vision and sometimes requires surgery. Scientists found that AI programs were pretty good at spotting which eyes would progress, with accuracy rates between 77-85%. The AI models worked best when they included information like eye shape measurements, patient age, and even lifestyle factors like eye rubbing. However, the research still needs more testing before doctors can confidently use these AI tools in their offices every day.

The Quick Take

  • What they studied: Can computer programs using artificial intelligence predict which people with keratoconus will experience worsening of their eye condition?
  • Who participated: The review analyzed 10 different research studies that together included 10,940 eyes from patients with keratoconus. These were people whose corneas were becoming cone-shaped, which affects vision.
  • Key finding: AI programs were moderately to highly accurate at predicting keratoconus progression, with success rates between 77.5% and 84.9%. The programs were especially good at spotting progression when they included information about eye shape, patient age, eye rubbing habits, and vitamin levels.
  • What it means for you: If you have keratoconus, AI tools may eventually help your eye doctor predict whether your condition will worsen, allowing for earlier treatment. However, these tools aren’t ready for everyday use in clinics yet—more testing is needed first.

The Research Details

This was a systematic review, which means researchers searched through medical databases (PubMed, Google Scholar, and ScienceDirect) to find all published studies about using AI to predict keratoconus progression. They looked for studies published in English that used computer learning methods to analyze how keratoconus changes over time.

The researchers carefully evaluated the quality of each study using a standard tool called QUADAS-2, which checks for bias and reliability. They then combined the findings from all 10 studies in a narrative summary, explaining what the research showed overall.

This approach is valuable because it brings together evidence from multiple studies, giving a broader picture than any single study could provide. It helps identify patterns and common findings across different research groups.

A systematic review is important because keratoconus is a serious eye condition that can lead to vision loss, and being able to predict who will get worse could help doctors treat patients earlier. By combining results from multiple studies, researchers can see which AI approaches work best and what information is most helpful for making predictions. This type of review also identifies gaps in the research that need to be filled before these tools can be used in real clinics.

The studies included were of moderate to good quality, but there are important limitations. Most studies only tested their AI programs on the same data they used to train them (called internal validation), rather than testing on completely new, independent patient groups. This is like a student studying only the exact practice problems that will be on the test—it doesn’t prove they’d do well on a different test. The studies also used different definitions of what counts as ‘progression,’ making it hard to compare results directly.

What the Results Show

The AI programs analyzed in these studies showed moderate to high accuracy in predicting keratoconus progression. The accuracy ranged from 77.5% to 84.9%, meaning the programs correctly predicted progression about 3 out of 4 times. When researchers measured how well the programs could distinguish between eyes that would progress and those that wouldn’t, they got scores (called AUC values) between 0.77 and 0.93, with higher numbers being better.

Three key factors consistently helped the AI programs make better predictions: the shape of the back of the cornea (posterior elevation), a measurement called maximum keratometry (Kmax) that shows how curved the cornea is, and the patient’s age. Importantly, people whose keratoconus got worse were significantly younger than those whose condition stayed stable.

When researchers added non-imaging information to the AI programs—such as IgE levels (related to allergies), eye rubbing habits, and vitamin D and B12 deficiencies—the programs became even better at predicting progression. This suggests that keratoconus progression isn’t just about eye measurements; lifestyle and nutritional factors matter too.

The research revealed that combining multiple types of information improved prediction accuracy. Eye rubbing, which is a known risk factor for keratoconus, was identified as an important factor that AI programs should consider. Vitamin deficiencies, particularly vitamin D and B12, appeared to play a role in disease progression. These findings suggest that managing these modifiable factors might help slow keratoconus progression, though more research is needed to confirm this.

This review builds on earlier research showing that AI and machine learning have potential in eye disease prediction. Previous studies suggested AI could be useful for eye conditions, and this review confirms that AI is promising for keratoconus specifically. However, this review also shows that earlier studies had similar limitations—most hadn’t been tested on completely independent patient groups, which is necessary before doctors can trust them in real-world practice.

Several important limitations affect how much we can trust these findings. First, none of the 10 studies tested their AI programs on a completely separate group of patients from a different hospital or clinic—they only tested on data they’d already used for training. Second, the studies used different definitions of what counts as keratoconus progression, making it hard to compare them fairly. Third, most studies relied on data from specific eye-testing machines, so the AI programs might not work as well with different equipment. Finally, the studies didn’t thoroughly report on how well the programs would work in real clinical settings, and they didn’t always explain their decision-making process clearly.

The Bottom Line

Current evidence suggests AI tools may help predict keratoconus progression, but they’re not ready for routine clinical use yet. If you have keratoconus, continue regular eye exams with your ophthalmologist. Ask your doctor about new screening tools as they become available. Managing modifiable risk factors—like reducing eye rubbing, maintaining adequate vitamin D and B12 levels, and managing allergies—may help slow progression, though more research is needed. Confidence level: Moderate—the tools show promise but need more rigorous testing.

This research matters most for people with keratoconus who want to know if their condition will worsen, and for eye doctors looking for better ways to monitor patients. It’s also relevant for researchers developing AI tools for eye diseases. People without keratoconus don’t need to worry about this research right now. If you have a family history of keratoconus or have been diagnosed with it, this information could be valuable for future care.

If AI tools are eventually approved for clinical use, they would likely provide predictions during regular eye exams. Benefits from managing risk factors like eye rubbing and vitamin levels might take weeks to months to show effects on disease progression. However, these tools are still in research stages, so widespread clinical availability is probably 2-5 years away.

Want to Apply This Research?

  • Track eye rubbing frequency (count instances per day), vitamin D and B12 supplement intake, and dates of eye exams. Record any changes in vision clarity or eye discomfort to discuss with your eye doctor.
  • Set reminders to reduce eye rubbing by using artificial tears when eyes feel itchy, take daily vitamin D and B12 supplements as recommended by your doctor, and schedule regular eye exams every 3-6 months if you have keratoconus.
  • Log weekly eye rubbing incidents, monthly vitamin supplement adherence, and maintain a record of all eye exam results including corneal measurements. Share this data with your eye doctor to help them monitor progression and adjust your care plan.

This research summary is for educational purposes only and should not replace professional medical advice. Keratoconus is a serious eye condition that requires ongoing care from a qualified ophthalmologist. The AI tools discussed in this review are still experimental and not yet approved for routine clinical use. If you have keratoconus or suspect you might, consult with your eye doctor about appropriate monitoring and treatment options. Do not make changes to your eye care routine based solely on this information without discussing it with your healthcare provider first.