New AI-based model may help identify patients at risk for post-LASIK ectasia
Click Here to Manage Email Alerts
A new AI-based model showed the ability to identify eyes with normal topographies at risk for developing post-LASIK ectasia.
“This method increases the number of cases correctly identified as at risk and reduces the number of eyes that had been inadequately considered at risk,” the authors wrote.
Source: Adobe Stock.
Six features, including percent tissue altered (PTA), residual stromal bed, corneal thickness, flap thickness, central ablation depth and age, were used to engineer through machine learning 14 additional features. The different interactions between these 20 variables were tested, sampling thousands of models with diverse predictive performance. Following fivefold cross-validation, the best performing model was selected. This included only two original variables, PTA and corneal thickness, and two variables derived from the engineering process that stood out as relevant: derivative PTA and age-weighted value (AWV).
This model is freely available and easy to use, Marcony R. Santhiago, MD, PhD, lead author of the study, told Healio/OSN.
“Readers interested in accessing it for free can contact me through my Instagram account @marconysanthiago. You simply enter the LASIK flap thickness (or specify if it is going to be PRK), ablation depth, corneal thickness and the patient’s age in years. We’ll do for you all the conversions and complex calculations,” he said.
The study included 65 eyes of 38 patients with normal bilateral preoperative Placido topography who developed ectasia after LASIK for myopia or myopic astigmatism and a control group of 274 eyes of 172 patients who did not develop post-LASIK ectasia over a postoperative follow-up of at least 5 years.
PTA derivative, PTA values and AWV were significantly higher, while corneal thickness was significantly thinner, in the eyes that developed ectasia as compared with controls. The AI-based model performed significantly better as compared with existing risk factors for identifying patients at higher risk. The method also gave fewer false-negative and false-positive errors.
“Our unprecedented modeling scale allowed us to find complex panels to estimate the conversion of apparently ordinary individuals into patients at higher risk for ectasia,” the authors wrote.
Collapse
Read more about
Click Here to Manage Email Alerts