Correlation of local glaucomatous damage in RNFL attenuation coefficient and thickness maps with visual field defects

Abstract

Purpose : Retinal nerve fiber layer (RNFL) thickness is the most commonly used optical coherence tomography (OCT) parameter to detect and monitor glaucomatous damage. Other OCT-derived parameters, such as attenuation coefficient (AC) values, may provide additional information about RNFL integrity. In this study, we visually determined the correlation between local glaucomatous damage in both RNFL AC and RNFL thickness maps and areas of decreased retinal sensitivity in visual field maps.

Methods : 20x20^o volumetric OCT scans (Spectralis, Heidelberg Engineering) and standard automated perimetry (HFA 24-2 SITA Standard, Carl Zeiss Meditec) tests were acquired in 56 glaucoma eyes. The obtained OCT data, consisting of 193 B-scans, was converted into ACs and the RNFL was segmented by custom software (Novosel et al, MedIA 2015). RNFL thickness and AC maps were created. On both of these maps, local structural defects were visually identified and annotated by a human grader (Fig.1a). Then, the annotations were transferred to a standardized template to ensure a masked grading (Fig.1b). Finally, the correlation between each defect on the templates and visual fields (VF) (Fig.1c) was graded as 1: probable, 2: possible, 3: questionable and 4: unlikely. The difference between the correlation of VFs with either AC or thickness maps was evaluated with the Wilcoxon Signed Ranks test.

Results : No statistically significant difference was found between the overall correlation scores of AC and thickness (p=0.77). Interestingly, AC and thickness scores disagreed in 24% (26/110) of the correlations (Fig.2b). Frequently, structural defects did not have a convincing corresponding VF defect, resulting in the high score for 4: unlikely (Fig.2c).

Conclusions : While the overall correlation scores of AC and thickness are very similar, they differ in a considerable number of cases and provide complementary information.

This is an abstract that was submitted for the 2016 ARVO Annual Meeting, held in Seattle, Wash., May 1-5, 2016.

 

Figure 1. RNFL maps overlaid with manual annotations (a), corresponding templates (b) and VF map (c). Case 1: Good correlation in 2 defects (green&blue) and poor correlation in 1 defect (red). Case 2: The blue defect in the thickness map correlates poorly with the VF.

 

Figure 2. Frequency histogram for AC and thickness (a). Confusion matrix showing the scores of AC in respect to thickness (b). C. Example with VF defects but no defects in the OCT maps. NA: not available, no correlation to be scored.