Abstract
Purpose :
To compare the performance of glaucoma risk score models developed by Fukai et al. (2022) to the UNC OCT Index model, developed by Mwanza et al. (2013) on healthy and diseased eyes scanned using a spectral domain optical coherence tomography (SD-OCT) device.
Methods :
A cross-sectional study was conducted on 25 glaucoma and 25 healthy patients. Patients underwent an ocular examination to determine if they belonged in the healthy or glaucoma group. Each patient was scanned by an SD-OCT device (Maestro, Topcon Healthcare, Tokyo, Japan) using 12 mm by 9 mm widefield OCT scans. Low-quality scans were excluded from the dataset for the following reasons: blinks, motion artifacts, a TopQ score of less than 25, segmentation errors, or a mispositioned optic disc. All the risk score models were applied to each of the OCT scans. The performance of the models was evaluated by generating area under the receiver operating characteristic curves (AUROCs), sensitivity, and specificity metrics.
Results :
The different models showed similar performance with AUROC values ranging from 0.92 to 0.96. Four of the five models achieved 100% specificity, accurately identifying healthy eyes without false positives. Model 3B from Fukai et al. had a higher AUROC than the UNC OCT Index by a difference of 0.04 and an 8% higher sensitivity for glaucoma detection.
Conclusions :
Both risk score models exhibited high performance in distinguishing healthy from glaucomatous eyes based on AUROC and specificity results. The UNC Model performed nearly as well as the models from Fukai et al. despite having been developed on a different OCT device. The ability of both of these models to perform well, with very high specificity, in this small independent test set, suggests that OCT-based scores have promise for detecting glaucoma in screening scenarios.
This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.