Artificial Intelligence-Derived Biomechanical Index for Early Corneal Ectasia Detection: Advancing Beyond Tomography.

PURPOSE: To develop and evaluate a novel artificial intelligence (AI)-derived metric for detection of early ectasia (E) by leveraging spatiotemporal biomechanical data derived from dynamic cornea videos only.

DESIGN

Multicenter cross-sectional case-control retrospective study.

PARTICIPANTS

A total of 451 eyes from 451 patients from 2 centers with both Scheimpflug tomography and dynamic corneal deformation examinations.

METHODS

The training data set included 167 normal (N) patients with stable eyes postlaser vision correction, 83 NT eyes with VAE (VAE-NT), and better eyes from 64 patients with bilateral keratoconus (ie, mild KC [MKC]). The external validation data set included 68, 33, and 36 patients from the N, VAE-NT, and MKC groups, respectively. Extensive spatial and temporal pixel-level analysis of corneal displacement, stromal gray scale changes, and thickness variations from Corvis ST video frames was performed to identify a novel risk score with best detection of early E (combined VAE-NT and MKC). Findings were further validated using various approaches based on the external data set.

MAIN OUTCOME MEASURES: Area under the receiver operating characteristics curve (AUC), accuracy, specificity, and sensitivity for detecting early E.

RESULTS

The novel AI-derived index achieved an AUC of 0.995 with accuracy, sensitivity, and specificity of 96%, 97%, and 96%, respectively, outperforming all Scheimpflug tomographic and dynamic deformation combined indices such as Belin/Ambrósio Enhanced Ectasia Display, Corvis Biomechanical Index (CBI), and Tomographic and Biomechanical Index (TBI).

CONCLUSIONS

The AI-derived index based on spatiotemporal corneal biomechanical data outperformed existing instrument's topographic, tomographic, and biomechanical indices, in detecting E. This novel metric offers a clinically meaningful enhancement in early diagnosis and management of corneal E disease.

FINANCIAL DISCLOSURES: Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.