Artificial Intelligence And Simulation Driven “Pathfinder” In Refractive Surgery

Presenting AcuSimX, the first-ever artificial intelligence (AI) and inverse finite element methods (iFEM) simulation based prediction tool for estimating postoperative (post-op) corneal stiffness after laser refractive surgery using preoperative (pre-op) data. We aimed to demonstrate the accuracy of AcuSimX while predicting corneal stiffness (CS) after small incision lenticule extraction (SMILE), laser assisted in situ keratomileusis (LASIK) and photorefractive keratectomy.

 

Narayana Nethralaya, Bengaluru, India

529 eyes of 529 patients were included in this study and were randomly divided into training (n = 371) and test (n =158) cohorts. Additionally, we also evaluated pre-op data of 12 eyes (10 SMILE and 2 LASIK eyes) that developed ectasia after surgery. In AcuSimX, an iFEM virtual patient specific corneal model was built from pre-op Corvis-ST deformation and Pentacam HR tomography (OCULUS Optikgerate Gmbh) data. iFEM used virtual model and planned aspheric ablation profile to estimate post-op CS.The computed post-op CS was refined using Lasso regression AI equation derived from training cohort and was validated on test cohort. An ectasia risk assessment nomogram was also built to classify the ectasia from normal eyes using a decision tree AI.

The mean absolute error was 6.24 N/m and the intraclass coefficient was 0.84 [95% confidence interval was 0.80-0.87] in the training cohort consisting of 371 eyes. Likewise, the mean absolute error was 6.47 N/m while the intraclass coefficient was 0.84 [0.78-0.89] in the test cohort consisting of 158 eyes. We also found that the actual post-op CS measured in-vivo and the software computed post-op CS was statistically similar (p > 0.05) in ectasia eyes. The ectasia risk assessment nomogram was able to classify all ectasia from normal eyes using pre-op input data and predicted post-op CS.  

Overall, an excellent intraclass coefficient (greater than 0.84) demonstrated the accuracy of predicted post-op CS. At present, there is no tool to predict the post-op biomechanical outcomes. The virtual surgery simulation platform AcuSimX is the first biomechanical simulation software for refractive surgery clinics which could serve as a “pathfinder” in choosing the right procedure. AcuSimX using predicted post-op CS and the ectasia risk assessment nomogram could help in avoiding ectasia completely.