Abstract
Purpose::
To develop progression analysis algorithms for GDx retinal nerve fiber layer (RNFL) measurements. To evaluate algorithm specificity based on short-term reproducibility data and to evaluate algorithm sensitivity through simulation test.
Methods::
Two modes of progression analysis have been developed for GDx RNFL measurements: Fast and Extended. Progression detection is based on analysis of 3 measurement groups: summary parameters, TSNIT plots, and RNFL images. Fast mode requires only one image per visit and is entirely based on change from baseline (CFB) method. Extended mode requires 3 images per visit and is based on statistical image mapping (SIM) method for summary parameters and TSNIT, and CFB method for RNFL image. CFB method is based on comparing the difference between later visits and baseline visits to the measurement reproducibility. SIM method is based on comparing the trend of the observed order with those of random permutations to determine the extremeness of the observed order. To evaluate the specificity, a short-term reproducibility study was conducted which included 14 volunteer subjects (28 study eyes), 2 operators, 5 GDx VCC instruments (Carl Zeiss Meditec Inc., Dublin, CA). Each subject was imaged on each instrument for 6 visits on 3 different days, two visits per day by the 2 operators. Each visit contained 1 cornea, 3 VCC, and 3 ECC images. To evaluate the sensitivity, a computer simulation program was developed to assess minimally detectable loss (at 90% detection rate) based on simulated data series with various reproducibility and type/amounts of loss.
Results::
Based on the short-term reproducibility data, the GDx progression analysis achieved the design goal of 95% specificity for VCC Fast mode, ECC Fast mode, and ECC Extended mode; the specificity for VCC Extended mode was approximately 90%. Based on simulation tests, approximately 10 steps across dynamic range was achievable for single instrument based data series; the Extended mode had slightly higher sensitivity than the Fast mode; summary parameters were more sensitive to diffuse loss with area greater than 60°, and TSNIT and RNFL image were most sensitive to localized loss of 20° ~ 40°; focal defect smaller than 10° was difficult to detect.
Conclusions::
The GDx progression analysis provides an objective tool to identify progressive RNFL loss in GDx measurements. Preliminary evaluation demonstrated reasonable specificity and sensitivity. Clinical evaluation is needed to assess the utility of the GDx progression analysis in the detection of glaucomatous progression.
Keywords: imaging/image analysis: clinical • nerve fiber layer • detection