The study of structure and function correlations in the macula by combining OCT imaging and VF assessment is thought to be important for understanding the nature of glaucomatous damage and for strengthening the basis for diagnostic decisions. Additionally, data from the macular testing can provide supplementary evidence to optic nerve head and retinal nerve fibre layer examination, particularly when a differential diagnosis has to be made, for example, in cases of high myopia.23–25 However, evidence to date has not provided a clear and unambiguous foundation for understanding how structural and functional damage are related to each other. In fact, methodologies of CMAs that examine this relationship have yielded at best moderate correlations with maximum correlation coefficients ranging from 0.41 to 0.77. 20–22,26−28 The present study was carried out to map the macular structure-function relationship by comparing conventional techniques to one employing fundus-tracking perimetry to accurately map each VF test location to corresponding locations in the retina to correlate VF sensitivity to GCL and IPL thickness values. However, our hypothesis that the LMA would yield higher correlations between VF sensitivity and OCT thickness values was not substantiated.
Our hypothesis was based on the assumption of accurate correspondence between the retinal location where the VF stimulus was projected and the retinal location where OCT measurements were made. We centred the GCL and IPL thickness values on each VF stimulus to ensure that measurements were localized. Based on a similar premise, but different methodologies, a comparable29 or even better30 structure-function relationship in the macula was reported with fundus-tracking perimetry. Contrary to these studies, our results showed that the correlation coefficients obtained with the LMA were significantly lower compared to the CMA.
Many previous reports on the structure-function relationship in the macula only explored the correlations between standard OCT sectors or superpixels and corresponding VF locations, constraining the correlations only to mapped locations to maintain the concordance between the anatomical distribution of RGCs and VF locations.20–22,26−28 In contrast, we explored correlations over the entire macula instead of exclusively between VF units and corresponding OCT superpixels. We found that most of the vectors that linked each VF unit to the OCT superpixel with the maximum correlation were longer than expected and directed towards retinal locations further from the theoretical mapped location. These findings were observed for both for the CMA (80% each for the GCL and IPL; Fig. 3 and Supplementary Fig. 1) and LMA (96% and 97% for GCL and IPL respectively; Fig. 4 and Supplementary Fig. 2).
Outside the central 8° of the macula, corresponding to approximately 2 mm from the fovea, the thickness of GCL and IPL decreases rapidly because the RGC somas are arranged in a single layer.1 Due to the normally lower GCL and IPL thickness in these areas, glaucoma is unlikely to result in further significant reductions in thickness. Hence, the GCL and IPL thickness in unaffected areas with corresponding normal VF sensitivities may be similar to those in areas damaged by glaucoma that have corresponding reduced VF sensitivity.31 Consequently, outside the 8° of the macula, very weak or even negative correlations were observed (Figs. 1 and 2), confirming previous findings.27,31 In contrast, within the central 8°, where the RGCs are stacked within multiple layers and where the GCL and IPL thickness is greater,1 glaucomatous damage produces progressive measurable thinning over a larger range of values. This progressive thinning, together with a corresponding reduction in VF sensitivity, yields a statistically higher correlation between structure and function. Lee and colleagues,27 who used a sector based analysis, also found that maximum correlations tend to be concentrated in areas with a wider range of measurements and where glaucomatous damage can be more readily detected.
The CMA used 3°x3° OCT superpixels, while the LMA used 2°x2° superpixels. The different resolution of the two approaches may further explain the unexpected higher correlation observed with the CMA. In the latter, a varying number of VF locations were used to derive VF units within which VF sensitivity was averaged. Furthermore, OCT superpixels were larger compared to the LMA where VF units were single VF locations and OCT superpixels were smaller. Averaging VF sensitivities, and GCL and IPL thickness over larger areas may have resulted in higher signal-to-noise ratio and consequently higher correlation coefficients with the CMA compared to the LMA. Additionally, the difference in the size of OCT superpixels and VF units between the two approaches could have affected the degree of error in vector direction and length. As result, we observed more vectors with spurious directions with the LMA compared to the CMA.
An additional factor that may explain the lower correlations obtained with LMA could be the greater experience subjects had with standard automated perimetry compared to fundus-tracking perimetry. Subjects were recruited from a prospective study in which they are regularly tested with standard automated perimetry. While unreliable tests with fundus-tracking perimetry were repeated when unreliable, we cannot rule out the possibility that learning effects or inadequate experience may have influenced the correlations obtained with the LMA.
For both the CMA and LMA, we noted more vectors directed temporally in the inferior retina (corresponding to the superior VF) compared to the superior retina (Figs. 3 and 4). The infero-temporal sector is considered to be susceptible and where damage is thought to occur early in the disease.2,32,33 These observations support the notion that the pattern of glaucomatous damage may drive the magnitude of maximum correlations and vector direction when studying the structure-function relationship. According to topographic models, nerve fibre bundles from the superior hemiretina take a more peripheral trajectory compared to those from the inferior hemiretina which course closer to the fovea.34,35 Therefore, damage to fibres in the superior retina could lead to a loss of RGCs further from the fovea and correspond to more peripheral inferior VF defects that may fall outside the central 10°. Conversely, damage to fibres in the inferior retina would correspond to superior VF defects closer to fixation inside the central 10°, and therefore could have led to a higher structure-function correlation inferiorly (Fig. 5), as also reported by others.22, 26–29,36,37
In most of the retina, photoreceptors are vertically aligned with the RGCs, however in the central 8 - 9°,1 because of the high concentration of cones and the foveal pit, corresponding RGCs are eccentrically displaced. For this reason, displacement models were conceived for more accurately mapping the projection of VF locations to RGCs corresponding to the stimulated photoreceptors.7,8 Theoretically, the maximum correlation between either GCL or IPL thicknesses and VF sensitivity should lie at the location identified by the displacement models. Additionally, studies using displacement models in structure-function analysis should yield higher correlations compared to those that do not use them. However, evidence to date indicates that application of displacement models fails to unequivocally demonstrate a significant increase in correlation.27,37,38 Indeed, our findings showed that the average length of the correlation vectors was considerably greater than the highest shift proposed by the models, which ranges from 1° to 3°.7,8,39
Our study is limited by the relatively small sample size, mainly due to the requirement of additional inclusion criteria that placed subjects into specific mean GCL thickness and 10-2 MD tertiles. We decided to add these criteria to ensure an adequate sample size to represent the range of glaucomatous damage. However, we acknowledge that inclusion of patients with very early glaucoma where measurements are not substantially different to age-matched control values, could have led to a weaker structure-function relationship.42 Likewise, in advanced glaucoma, correlations may be inaccurate because the lower end of the measurement range of the perimeters and OCT devices may be reached at different stages of the disease.20,41 Hence for a given patient, the veracity of VF and OCT measurements could be variable.
In summary, contrary to our hypothesis, the LMA, using fundus-tracking perimetry to accurately map VF locations to corresponding GCL and IPL thickness values, did not improve the structure-function correlations in the macula. The poor correlation of the CMA is less likely due to inaccurate mapping of VF locations to the retina, but more likely due to factors such as variability in measurements, which affect both forms of perimetry and OCT and specific patterns of localized damage that drive vector direction and strength of maximum correlations. Clinical judgment that subjectively relates structural and functional losses in the macula in relatively large sectors, for example, decreased GCL and IPL thickness to superior VF loss remains useful. However, given the findings of this study, it is unclear how additional efforts to exploit macular structure-function relationships in an objective and quantitative manner can aid clinical decision making in glaucoma.