Abstract
Purpose:
Maps relating the visual field (VF) to the optic nerve head (ONH) have typically been produced by hand-tracing of nerve fibre bundles in retinal images by single observers. We have published a computational model incorporating biometric parameters as an alternative approach (Denniss et al, IOVS 2012). Here we investigate conviction and variability in nerve fibre tracing, and compare maps produced to the model. We hypothesised that discordance between the methods would be related to uncertainty and variability in tracing.
Methods:
VF locations (n=52, 24-2 pattern) were overlaid on composite 490nm scanning laser ophthalmoscope (F-10, Nidek, Japan) retinal images from 10 subjects with varied anatomy (axial length, ONH position). Ophthalmically trained (n=8) and untrained (n=5) observers viewed the images under standardised conditions and manually traced the 1° ONH sector into which nerve fibres from each VF location entered. Observers also recorded the range of sectors into which they were certain the fibre entered. These ranges were scaled for each observer/image to represent conviction in tracing. Traced sectors were compared to 10° sectors predicted by the computational model based on axial length and ONH position.
Results:
Across VF locations, variability in tracing fibres was high with median difference between two furthest apart traced sectors for a VF location of 27° (IQR 20-38°) for trained observers and 38° (IQR 28-56°) for untrained observers. Across all traced VF locations, concordance between the model and the range within which observers reported they were certain the fibre entered was 53% for trained and 79% for untrained observers. Across VF locations, conviction in tracing decreased linearly with distance from the ONH (trained: R2=0.81, untrained: R2=0.78, both p<0.001). Conviction was inversely correlated with tracing variability (both groups Spearman’s rho=0.66, p<0.001) and root mean squared differences between traced and model-predicted sectors (Spearman’s rho, trained 0.47, untrained 0.44, both p<0.001).
Conclusions:
Concordance between traced and model-predicted sectors was moderate, and greater for locations where conviction in tracing was greater. Variability and uncertainty in tracing nerve fibre bundles, particularly further from the ONH, should be taken into account when used to create or as a reference for structure-function maps.
Keywords: 758 visual fields •
627 optic disc •
610 nerve fiber layer