advertisement

Editors Selection IGR 11-4

Comments

Comment by David Garway-Heath on:

48060 Retinal Ganglion Cell Layer Thickness and Local Visual Field Sensitivity in Glaucoma, Raza AS; Cho J; De Moraes CGV et al., Archives of Ophthalmology, 2011; 129: 1529-1536

Find related abstracts

The issue raised by Raza et al. is whether putting structure and function data together provides more diagnostic information that either does separately. This paper shows how far we still must go. Every resident learns that inferior rim notches should correlate to an upper field defect. The Garway-Heath map was a valiant attempt to match near-disc image data to fields, but it is difficult to know exactly where the ganglion cells (or their axons) are that correspond to a zone of decreased field sensitivity from disc topography or peri-rim NFL thickness. The problems include a high degree of variability in number of RGCs per retina among humans (we range from 500,000 to 3 million per eye) and variety in disc orientation and size, as well as variations with axial length.

Retinal layer segmentation in SDOCT of RGC and inner plexiform layers (but not the NFL) was matched to the 10-2 field test point sensitivity corresponding to that retinal area (with correction for RGC displacement nearest fovea). As seen with earlier OCT, however, there is a wide range of retinal thicknesses corresponding to near normal function, as well as a floor effect of 45 microns of measured retina that isn't RGCs. The retinal layer thickness bottoms out at -10 dB, so it's adding nothing beyond that to the field. There are very low retinal thickness values with total deviation < 5 dB. This either means that there is huge variation in retinal thickness with normal function, or, that retinal thickness falls quite considerably prior to field loss (both are probably true). But, do we really want structure and function to correlate? If both are locally abnormal, their information might be additive. Or, it could be merely complementary. But, if the thickness changes before the field, then poor correlation is 'good'. A newer analytic framework is needed to see if this kind of comparison of two data sets adds to our predictive power.

• ← Previous
• Next →