A Novel Optical Coherence Tomography-based Keratoconus Diagnostic Index Incorporating Stromal and Epithelial Features.

PURPOSE: To establish a diagnostic index for keratoconus based on spectral-domain optical coherence tomography (SD-OCT) and to compare it with existing parameters.

METHODS

SD-OCT and Scheimpflug-based tomography were conducted on normal and keratoconic eyes. Multiple SD-OCT machine-derived parameters were assessed for the whole cornea, stroma, and epithelium. Receiver operating characteristic (ROC) curves were performed to determine area under the curve (AUC), sensitivity, and specificity. Principal component analysis and multinomial logistic regression after features selection established a new diagnostic index (Whole Information of Stroma and Epithelium [WISE]). The WISE index was compared with existing Scheimpflug-based diagnostic parameters.

RESULTS

A total of 306 healthy control, 101 forme fruste keratoconus (FFKC), 86 early keratoconus (EKC), and 161 advanced keratoconus eyes were included for training and internal validation, as well as 52 normal, 31 FFKC, and 36 EKC eyes as a test dataset. The highest-ranked SD-OCT parameters to discriminate FFKC and EKC from normal eyes were Pachymetry_9mm_N (AUC = 0.65) and Epithelium_5mm_SN-IT (AUC = 0.77). In the internal validation and test datasets, the proposed WISE index demonstrated AUC = 0.76 and 0.83 for FFKC, and = 0.92 and 0.94 for EKC, respectively, comparable to Belin-Ambrósio Deviation and Pentacam Random Forest Index, as confirmed by De-Long's test (All P > .10).

CONCLUSIONS

Individual OCT-based machine-derived parameters lack sufficient power to discriminate FFKC and EKC from normal corneas, but this can be improved by combining OCT-based information from stroma and epithelium as in this new index. The discrimination accuracy of the WISE index was comparable to existing Scheimpflug-based indices.