An automated, structure-function method for detecting progression of glaucoma

Abstract

Purpose : To develop and test an automated method for detecting progression based upon topographical agreement of changes in the optical coherence tomography (OCT) maps and 24-2 and 10-2 visual fields (VF) in eyes with early glaucoma.

Methods : Widefield OCT (12x9mm) scans and 24-2 and 10-2 VFs were acquired from 55 eyes with early or suspected glaucoma (baseline 24-2 mean deviation >-6dB) and 25 healthy controls (HC) as part of a longitudinal, prospective study. Baseline (V1) OCT maps and VFs were compared to those from the last follow-up visit (V2), which was obtained at least 1 year later. Continuous probability change (pc-) maps were determined for both the retinal nerve fiber layer (RNFL) and retinal ganglion cell plus inner plexiform layer (GCL+) (Fig. 1) by comparing the difference of the thickness between V1 and V2 to an estimation of short-term variability for scans within 4 months. Red (p<1%) and yellow (p<5%) in Figs. 1 & 2 indicate significant change. The same approach was applied for the VFs. A custom R program [1] superimposed VF locations on RNFL and GCL+ pc-maps and determined the number of locations with both progressive structural (pS, p<10%) and functional (pF, p<5%) change (diamonds in Figs. 1&2). The criterion number of pS-pF locations for a 95% specificity was estimated based upon sampling the HC group with replacement. A reference standard based upon experts’ evaluation of V1 and V2 OCT data (b-scans, and RNFL and GCL thickness and change maps) and VF tests identified 12 progressing (P) eyes.

Results : For the HC eyes, a threshold of 10 total pS-pF locations on the RNFL and GCL+ pc-maps achieved a specificity of 95% (1 HC false positive). The same threshold identified 8 eyes as ‘progressors’, 7 of which were true positives (sensitivity from 12 P eyes: 58%). A post-hoc analysis of the other 5 P eyes revealed that 4 of the 5 showed a very narrow progressive region on the OCT with pS-pF agreement within the same arcuate region, but no overlap (Fig. 2). The commercial GPA correctly identified only 4 of the 12 P eyes (sensitivity: 33%).

Conclusions : An automated method for detecting progression based upon topographical OCT and VF agreement showed good specificity and identified more P eyes than the commercial GPA. Eyes with narrow arcuate progression can still be missed, mainly due to the spatial resolution of the 24-2 and the limited extend of the OCT scan size. 1. Tsamis et al, TVST, 2020.

This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.