Abstract
Purpose :
To develop and test an automated method for topographically comparing abnormal regions on optical coherence tomography (OCT) and visual field (VF) probability maps in glaucoma suspect eyes or eyes with early glaucoma
Methods :
90 eyes from the MAPS database with a referring physician diagnosis of: Definitely Glaucoma (DG; n=53), Probably Glaucoma (PG; n=17), or Not Glaucoma (NG; n=20) along with 45 healthy eyes (total n=135; one eye per patient) were studied with OCT and VF. A custom R program was developed which detects overlap by superimposing 24-2 and 10-2 (SITA-Standard) VF on retinal nerve fiber (RNF) and retinal ganglion cell plus inner plexiform layer probability plots from widefield (12x9 mm; Topcon Atlantis) OCT scans. The number of locations with abnormal structure (aS) and function (aF) agreement was calculated for various combinations of aF (≤ 1, 2 and 5%) and aS (≤1, 5, 10 and 15%). Logistic regression was used to derive ROC curves and define optimal number of locations for each criterion. The best performing criteria, based on area-under-the-curve (AUC) measures, were compared on unrotated vs. rotated (based on fovea-to-disc angle) OCT scans and on VF total (TD) vs. pattern deviation (PD) probability maps
Results :
VF PD maps with a probability <5% and rotated OCT maps with a probability <15% had the best AUC (0.92). VF TD maps with a probability <5% and a rotated OCT map with a probability <10% performed almost as well (AUC: 0.91). The optimal number of abnormal locations was found to be 2, for both sets. At optimal level, the PD and TD criteria had an accuracy of 0.88 and 0.84, respectively. Positive likelihood ratio (LR) was 5.94 for the PD and 7.30 for the TD criteria, while negative LR was 0.10 and 0.24 respectively. When applied to unrotated OCT scans, the accuracy of PD and TD criteria decreased to 0.82 and 0.80, respectively
Conclusions :
Abnormal regions on OCT and VF probability maps can be topographically compared with an automated method. The automated method demonstrated that rotated scans have higher accuracy compared to unrotated scans, whereas VF TD and PD probability maps can have equal performance, albeit using different criteria
This abstract was presented at the 2019 ARVO Annual Meeting, held in Vancouver, Canada, April 28 - May 2, 2019.