Author: Giovanni Montesano (United Kingdom)
Co-authors: Bethany E. Higgins, Hannah Dunbar, Maximilian Pfau, Steffen Schmitz-Valckenberg, Jan H. Terheyden, David P. Crabb
Integrating structural and functional data is of key importance to characterise eye diseases. High resolution structural data can be acquired with spectral domain optical coherence tomography (SD-OCT), which allows accurate segmentations and measurement of different retinal layers. Functional maps of retinal sensitivity can instead be obtained with microperimetry, which incorporates live retinal tracking technology into standard perimetry to account for eye movements during the test. Both microperimetry and SD-OCT have been widely used to study age-related macular degeneration (AMD). Structural and functional changes in AMD are mainly thought to involve the outer retina. However, retinal ganglion cells (RGC) can also experience degeneration either from normal ageing or other diseases, such as glaucoma. RGC loss, especially when mild, can be partially gauged with SD-OCT and neural models provide a connection between estimates of stimulated RGCs and expected perimetric sensitivity. The purpose of this analysis is to use OCT-derived RGC estimates, within an established neural model for perimetry, to predict the expected sensitivity based on the structure of the inner retina. Deviations from the model, in patients with AMD, may indicate outer retinal damage in the absence of other pathology.
European multi-centre, low-interventional cohort study of age-related macular degeneration (MACUSTAR study) across seven European countries.
Microperimetry was performed using the MAcular Integrity Assessment device (MAIA, CenterVue, Padua, Italy). Both mesopic and scotopic tests (with a red stimulus) were performed on subjects with no AMD (controls), early AMD (eAMD) and intermediate AMD (iAMD). Scotopic testing was preceded by 30 minutes of dark adaptation. One eye of each participant was tested twice in separate sessions (within three weeks of each other). Dense macular scans (30°x25°, 241 B-scans) were obtained by Spectralis SD-OCT (Heidelberg Engineering, Heidelberg , Germany). Layer segmentations were obtained from a pre-trained and validated deep learning algorithm (based on DeepLab V3 architecture). Histology data were combined with Ganglion Cell Layer (GCL) thickness to obtain local structural estimates of RGC counts. Fundus images from the MAIA and the Spectralis were matched to accurately report the locations tested with perimetry onto the structural map. RGC displacement was applied to perimetric stimuli to measure the number of RGCs stimulated at each location. A bivariate normative linear mixed-effect model, accounting for both the local log10(RGC count) and age, was estimated for scotopic and mesopic microperimetry using the control group. Pointwise deviations from this model were quantified for the eAMD and iAMD group.
The final dataset included data from (mesopic/scotopic): 43/41 eyes with no AMD (controls), 28/27 with early AMD (eAMD) and 150/151 with intermediate AMD (iAMD). The mean (standard deviation) age (years) was 69 (7) for controls, 72 (6) for eAMD and 71 (7) for iAMD. The relationship between log10(RGC count) and sensitivity was significant (p < 0.001) for both the mesopic and the scotopic test. In agreement with previous results, the coefficients were compatible with partial summation conditions, which justifies the use of a linear relationship. However, the relationship was much shallower for the scotopic test (1 dB/ log10(RGC count)) than the mesopic test (1.72 dB/ log10(RGC count)). No significant effect of age was found (p = 0.319 and 0.673 for the mesopic and scotopic tests respectively). Both eAMD and iAMD eyes showed a significant deviation from the normative model (p < 0.05 for both scotopic and mesopic tests), but there was no significant difference between the two groups. Both tests identified a similar percentage of abnormal locations (sensitivity < 5th normative percentile) in eAMD (18% mesopic; 17% scotopic) and iAMD (17% for both tests) eyes, with no significant differences.
Our analysis, in agreement with previous findings, shows that the expected retinal light sensitivity can be estimated from a healthy inner retina assuming partial summation conditions. This is valuable for AMD because it offers a generalisable structural benchmark for the expected normative sensitivity. The approach accounts for inner retinal changes due to normal ageing as well as any pathology, which are frequent in an elderly population (e.g., undiagnosed/early glaucoma). Such a relationship, however, is not expected to hold for advanced RGC loss, both because of the shortcomings of structural estimates and because of deviations from the partial summation regime. Further investigation is required for a more harmonious integration of advanced inner retinal damage, such as from glaucoma. Eyes with AMD showed significant deviation from the normative models, but no significant differences between eAMD and iAMD groups. Of note, the iAMD group showed a wider variation in sensitivity, which might be due to the much larger sample size. Further analyses focussed on more extreme normative cut-offs and on locations with detectable structural outer retinal changes might be helpful in identifying finer differences between eAMD and iAMD.
Consultant - CenterVue, SpA