Corvis^ST biomechanical indices in the diagnosis of corneal stromal and endothelial disorders: an artificial intelligence-based comparative study.

Abstract

To analyse the value of the Corvis^ST indices in diagnosing corneal stromal and endothelial disorders (CSEDs). This institutional retrospective case-control study included 903 eyes with a CSED and 597 normal eyes (controls), assessed with Corvis^ST and MS39. Corvis^ST indices. The collected data were divided into a training set (70%) and a test set (30%). Artificial intelligence frameworks were used to distinguish each disorder from controls and to classify corneas into seven groups: keratoconus, high-risk corneas for keratoconus, laser corneal refractive surgery (LCRS), endothelial disorders, stromal opacities, glaucoma corneas and normal corneas. Stress-strain index (SSI) significantly increased with age in the control group. Compared with controls matched for age/sex, keratoconus was associated with Corvis Biomechanical Index (CBI) >0.51 (area under the curve, 0.99), Ambrósio's relational thickness horizontal (ARTh) <425.5 (0.97), deflection amplitude at the time of the first applanation (SPA-A1) <96.3 (0.97) and Pachy<522.4 µm (0.91); high-risk corneas with a difference in CBI between fellow eyes (CBI SYM) >0.14 (0.98), (L2) <1.95 (0.83) and Pachy<549.7 µm (0.71); LCRS with ARTh<455.1 (0.93) and CBI>0.35 (0.83); corneal endothelial disorders with Pachy SYM>19.7 µm (0.83), Pachy>569.1 µm (0.82) and CBI SYM>0.14 (0.77); stromal opacities with SPA-A1 SYM>11.8 (0.92), ARTh<569.9 (0.89), SSI SYM>0.14 (0.89) and CBI>0.22 (0.86). A logistic regression function using all indices reached an area under the receiver operating characteristic curve of 0.81 for glaucoma diagnosis. The TabPFN model provided the best accuracy (88.7%) for diagnosing the seven corneal conditions. SSI, SPA-A1, CBI and Pachy correlated with keratoconus grade. Keratoplasty for keratoconus improved but failed to restore normal corneal biomechanics. Corvis^ST indices are relevant for diagnosing CESDs and distinguishing various disorders from each other.