Scatterplot showing the associationbetween the scanning laser polarimetry parameter Nerve Fiber Indicator andthe global retinal nerve fiber layer (RNFL) photographic severity score.
Receiver operating characteristiccurves for the scanning laser polarimetry parameter Nerve Fiber Indicator(NFI) and for the global retinal nerve fiber layer (RNFL) photographic scoreto discriminate glaucomatous from healthy eyes.
Sensitivity and specificity valuesfor each cutoff point of the scanning laser polarimetry parameter Nerve FiberIndicator to discriminate glaucomatous from healthy eyes.
Medeiros FA, Zangwill LM, Bowd C, Mohammadi K, Weinreb RN. Comparison of Scanning Laser Polarimetry Using Variable Corneal Compensationand Retinal Nerve Fiber Layer Photography for Detection of Glaucoma. Arch Ophthalmol. 2004;122(5):698-704. doi:10.1001/archopht.122.5.698
Copyright 2004 American Medical Association. All Rights Reserved.Applicable FARS/DFARS Restrictions Apply to Government Use.2004
To compare retinal nerve fiber layer (RNFL) measurements obtained withscanning laser polarimetry (SLP) using variable corneal polarization compensationwith standard red-free photography for detection of RNFL damage in glaucoma.
This observational, cross-sectional study included 1 eye of each of42 patients with open-angle glaucoma, 32 patients suspected of having glaucoma,and 40 healthy subjects. The RNFL measurements using SLP with variable cornealcompensation were obtained within 3 months of red-free photographs. Two independentobservers graded RNFL photographs using a standardized protocol. Superiorand inferior hemiretinas were scored separately, and a global score was obtainedby averaging scores from each hemiretina.
Main Outcome Measures
The RNFL photography scores were compared with RNFL thickness measurementsobtained with SLP. The receiver operating characteristic (ROC) curves wereconstructed to assess the abilities of the different methods to differentiateglaucoma patients from healthy subjects.
The RNFL thickness decreased with increased RNFL damage as assessedby photographs in both hemiretinas (R2 =15%-47%). The area under the ROC curve for the best SLP parameter, Nerve FiberIndicator, was significantly greater than the area under the ROC curve forthe global RNFL photography score (0.91 vs 0.84, P =.03).
A moderate correlation was found between RNFL thickness measurementsobtained with SLP and RNFL scores from red-free photographs. Compared withsemiquantitative RNFL photography scores, the best SLP parameter had a higherdiagnostic accuracy to separate glaucoma patients from healthy subjects.
Retinal nerve fiber layer (RNFL) abnormalities have been shown to precedethe development of visual field defects in glaucoma patients.1- 3 AlthoughRNFL photographs have been established as a standard method for detectionof RNFL defects, the qualitative and subjective nature of this assessment,as well as the requirement for maximal pupillary dilation and optimal mediaclarity, limits its widespread applicability.4
Scanning laser polarimetry (SLP) is a diagnostic tool developed to quantitativelymeasure the thickness of the peripapillary RNFL. It is based on the principlethat polarized light passing through the birefringent RNFL undergoes a measurablephase shift, known as retardation, which is linearly related to histologicallymeasured RNFL tissue thickness.5 Although differencesin retardation between healthy and glaucomatous eyes have been previouslydescribed,6- 11 earlierstudies12,13 evaluating RNFL photographsand SLP in glaucoma showed only modest associations between these 2 methods.
A significant source of error in previous studies evaluating the SLPtechnology was most likely introduced by the erroneous compensation of anteriorsegment birefringence.14- 19 TheRNFL is not the only birefringent structure in the eye. The cornea and Henlefiber layer of the macula, and to a lesser extent the lens, also are birefringent.20- 23 Toaddress anterior segment birefringence, the first commercial SLP had a fixedcorneal compensator. The compensator was calibrated based on the assumptionthat all individuals had a slow axis of corneal birefringence 15° nasallydownward with a magnitude of retardance of 60 nm. However, there is a widevariation in both the axis and the magnitude of corneal birefringence in healthyand glaucomatous individuals.17,18,22,24 Animprovement of the SLP technology consisting of the variable compensationof anterior segment birefringence has been recently described.25 Weinrebet al26 showed that SLP with variable compensationof the axis and magnitude of corneal birefringence results in improvementof the sensitivity and specificity of several parameters to discriminate betweenhealthy and glaucomatous eyes. Further, monkey studies27,28 havesuggested that RNFL measures obtained with SLP using variable corneal compensation(VCC) better reflect the RNFL qualitative appearance than measures obtainedusing SLP with fixed corneal compensation.
To determine the clinical validity of any new technology, it is importantto compare it with previously established diagnostic techniques with provenclinical utility. Hence, the purpose of this study was to compare RNFL measurementsobtained with SLP using VCC to standard red-free photography for detectionof RNFL damage in glaucoma.
This observational, cross-sectional study included 114 eyes of 114 patients.Forty-two patients had open-angle glaucoma, 32 patients were suspected ofhaving glaucoma, and 40 were healthy. Mean ± SD age of glaucoma patients,patients suspected of having glaucoma, and healthy individuals was 67 ±11 years, 61 ± 12 years, and 65 ± 11 years, respectively (P = .06, analysis of variance). All patients were evaluatedat the Hamilton Glaucoma Center, University of California, San Diego, andretrospectively selected from our research database. These patients were includedin a prospective, longitudinal study designed to evaluate optic nerve structureand visual function in glaucoma. All patients who met the inclusion criteriadescribed were enrolled in the current study. Informed consent was obtainedfrom all participants. The University of California, San Diego, Human SubjectsCommittee approved all protocols, and the methods described adhered to thetenets of the Declaration of Helsinki.
Each subject underwent a comprehensive ophthalmologic examination, includingreview of medical history, best corrected visual acuity, slitlamp biomicroscopy,intraocular pressure (IOP) measurement using Goldmann applanation tonometry,gonioscopy, dilated funduscopic examination using a 78-diopter (D) lens, stereoscopicoptic disc photography, and automated perimetry using 24-2 full-thresholdstandard automated perimetry or Swedish Interactive Threshold Algorithm (Zeiss-Humphreyfield analyzer, Zeiss-Humphrey, Dublin, Calif). To be included, subjects hadto have best corrected visual acuity of 20/40 or better, spherical refractionwithin ±5.0 D, cylinder correction within ±3.0 D, and open angleson gonioscopy. Eyes with coexisting retinal disease, uveitis, or nonglaucomatousoptic neuropathy were excluded from this investigation. One eye of each patientwas randomly selected for inclusion in the study.
The eyes of the healthy controls had IOPs of 22 mm Hg or less, withno history of increased IOP and a normal visual field result. Normal visualfield was defined as a mean deviation (MD) and pattern standard deviation(PSD) within 95% confidence limits and Glaucoma Hemifield Test results withinnormal limits. Healthy control eyes also had a healthy appearance of the opticdisc and RNFL (no diffuse or focal rim thinning, cupping, optic disc hemorrhage,or RNFL defects), as evaluated by optic disc photographs.
Eyes were classified as glaucomatous if they had repeatable (2 consecutive)abnormal visual field test results, defined as a PSD outside the 95% normalconfidence limits or Glaucoma Hemifield Test results outside 99% normal confidencelimits, regardless of the appearance of the optic disc. Average MD of theglaucomatous eyes on the visual field test nearest the imaging date was −4.92dB. According to the Hodapp-Parrish-Anderson29 gradingscale of severity of visual field defects, 27 patients (64%) were classifiedas having early visual field defects, 9 patients (21%) had moderate defects,and 6 patients (14%) had severe visual field defects.
Patients suspected of having glaucoma had either ocular hypertension(IOP >22 mm Hg on more than 2 separate visits) or glaucomatous appearanceof the optic disc but normal results on visual field tests. Glaucomatous damageto the optic disc was defined as the presence of neuroretinal rim thinning,excavation, notching, or characteristic RNFL defects. Of the 32 patients suspectedof having glaucoma with normal visual fields, 12 (38%) had ocular hypertensionand normal optic nerves, whereas 20 (62%) had a glaucomatous appearance ofthe optic disc.
The RNFL photographs were acquired with a TRC-50VT camera (Topcon AmericaCorp, Paramus, NJ) using Kodak Kodalith high-contrast film (Eastman KodakCompany, Rochester, NY). Superior and inferior 20° fields were obtainedusing a set of narrow-pass filters (SE-40 and SE-50, Spectrotech Filter, Saugus,Mass) and a standard Topcon red-free filter. In addition, 55° wide-angleviews of the optic disc centered between the superior and inferior regionsof the RNFL were obtained. Photographs were printed on 35-mm transparenciesand viewed on a light box with an acrylic optical grade magnifier (originalmagnification, ×4).
Evaluation of the RNFL photographs was completed using a semiquantitativemethod described by Niessen et al.30 In thissystem, photographs were compared with a set of 25 reference photographs,and a point scale was used to score the photographs according to nerve fibervisibility.30,31 A high scoreof 25 indicated normal, thick, striated fibers, whereas a low score of 1 indicatedthat no nerve fibers were visible. Superior and inferior arcuate bundles werescored separately. Each photograph was assessed by 2 graders (F.A.M. and K.M.),and the final RNFL score was calculated as the average of the RNFL scoresof each grader. Each grader was masked to the subject's identity and the othertest results. Disagreements (difference of more than 4 points between the2 graders) were resolved by consensus of adjudication of a third grader (C.B.).A global severity score (RNFL global) for each eye was created by averagingthe superior and inferior final RNFL scores. Poor-quality RNFL photographs,as assessed by consensus of the graders, were discarded. Seventeen patients(13%) of an initial group of 131 eligible patients had poor-quality RNFL photographsand were excluded from the study, leaving 114 subjects for the analysis.
All patients underwent imaging with a commercially available SLP (GDxVCC, software version 5.0.1; Laser Diagnostic Technologies Inc, San Diego,Calif) within 3 months of RNFL photographs. The general principles of SLPhave been described in detail elsewhere.8 TheGDx VCC is a modified SLP system with VCC. Images of the ocular fundus areformed by scanning the beam of a near infrared laser (780 nm) in a rasterpattern. The scan raster covers an image field 40° horizontally and 20°vertically in the eye, covering both the parapapillary and the macular regions.32 In the GDx VCC, the method of VCC, as described byZhou and Weinreb,25 has been automated andreplaced the original fixed corneal compensator. The VCC in the system consistsof 2 identical linear retarders in rotating mounts so that both the retardanceand axis of the unit can be adjusted according to requirements. To measureeye-specific corneal polarization axis and magnitude, SLP images of the maculaare first acquired without compensation (the retardance of the VCC is setto zero). The radial birefringence of the Henle fiber layer in the maculais used as an intraocular polarimeter, and both the Henle fiber layer andcorneal retardance can be determined from the macular retardation profile.Next, corneal birefringence-compensated SLP images are obtained using theappropriate eye-specific corneal polarization axis and magnitude values byadjusting the VCC retarders. The GDx VCC measures retardation in units ofnanometers. To simplify communications, retardation values are converted intothickness values (micrometers) using a fixed conversion factor of 0.67 nm/µm.32
A baseline image was automatically created from 3 images obtained foreach subject and used in each analysis. Only scans of good quality were included.Quality assessment was evaluated by an experienced examiner masked to thesubject's identity and results of the other tests. Good-quality image requireda focused and evenly illuminated reflectance image with a centered optic disc.Six patients (5%) of an initial group of 131 eligible patients had poor-qualitySLP images, and all of them had also poor-quality RNFL photographs.
The GDx VCC software calculates summary parameters based on quadrants,which are defined as temporal (335°-24° unit circle), superior (25°-144°),nasal (145°-214°), and inferior (215°-334°). Both hemifieldand global parameters were evaluated. The hemifield SLP parameters investigatedin this study were superior ratio (superior quadrant thickness to temporalquadrant thickness), inferior ratio, superior-nasal ratio, superior maximum(average of the 1500 thickest points in the superior quadrant), inferior maximum,superior average, inferior average, normalized superior area (area under theTSNIT [temporal-superior-nasal-inferior-temporal] curve in the superior quadrant),and normalized inferior area. Global SLP parameters investigated were maximummodulation ([thickest quadrant − thinnest quadrant]/thinnest quadrant),ellipse modulation, ellipse average, ellipse standard deviation, and the NerveFiber Indicator (NFI). The NFI is calculated using a support vector machinealgorithm based on several RNFL measures (Michael Sinai, PhD, Laser DiagnosticTechnologies Inc, written communication, March 2003) and assigns a numberfrom 0 to 100 to each eye. The higher the NFI, the greater the likelihoodthe patient has glaucoma.
Analysis of variance was used to assess differences in SLP parametersand RNFL photography scores among glaucoma patients, patients suspected ofhaving glaucoma, and healthy individuals. The Fisher least-significant differencetest was used to perform post hoc multiple comparisons. The correlation betweenthe semiquantitative RNFL severity scores obtained by assessment of RNFL photographsand the SLP parameters was evaluated by Pearson product moment correlationcoefficients. The RNFL scores for each hemiretina were compared with SLP parametersevaluating the corresponding hemiretina, whereas global RNFL scores were comparedwith global SLP parameters.
Receiver operating characteristic (ROC) curves were used to describethe ability to differentiate glaucomatous from healthy eyes of each GDx VCCsoftware–provided parameter and also of the RNFL photographic scoringsystem. The ROC curve shows the tradeoff between sensitivity and 1 −specificity. An area under the ROC curve of 1.0 represents perfect discrimination,whereas an area of 0.5 represents chance discrimination. The method of DeLonget al33 was used to compare areas under theROC curve. Minimum specificity cutoffs of 80% and 95% were used for comparingthe sensitivity of the SLP parameters and the RNFL photographic scoring method. P<.05 was considered statistically significant. Statisticalanalyses were performed using SPSS statistical software, version 10.0 (SPSSInc, Chicago, Ill).
Table 1 gives the mean valuesof all SLP parameters and RNFL scores in glaucoma patients, patients suspectedof having glaucoma, and healthy individuals. Significant differences betweenglaucoma patients and healthy subjects and between glaucoma patients and patientssuspected of having glaucoma were found for all SLP parameters and also forRNFL photographic scores. When patients suspected of having glaucoma werecompared with healthy individuals, the SLP parameters NFI, superior maximum,and normalized superior area showed statistically significant differencesbetween the 2 groups.
Table 2 gives the correlationcoefficients for the associations between SLP parameters and RNFL photographicscores. All correlations were statistically significant with P<.001. Lower RNFL scores were associated with thinner RNFL thicknessmeasurements as indicated by the SLP parameters. For the superior hemiretina,correlations between superior RNFL scores and corresponding SLP parametersshowed R2 values ranging from 16% to 40%.For the inferior hemiretina, the R2 valuesranged from 16% to 37%. For the association between global RNFL scores andglobal SLP parameters, the R2 values rangedfrom 15% to 47%. The strongest relationship was found between the SLP parameterNFI and the global RNFL score (r = −0.683, R2 = 47%, P<.001,Pearson correlation coefficient). As the SLP parameter NFI increased (moredamage), the global RNFL photograph severity score decreased (more damage)(Figure 1).
We also investigated the correlation between the SLP parameters andRNFL scores with the visual field indices MD and PSD. The SLP parameter NFIshowed the best correlation with MD and PSD among the SLP parameters. Highervalues of NFI were associated with lower values of MD (r = −0.618, R2 = 38%, P<.001, Pearson correlation coefficient) and highervalues of PSD (r = 0.597, R2 = 36%, P<.001, Pearson correlation coefficient).For the RNFL photographic scoring system, lower values of the global RNFLscore were significantly associated with lower MD values (r = 0.565, R2 = 32%, P<.001, Pearson correlation coefficient) and higher PSD values (r = −0.526, R2 =28%, P<.001, Pearson correlation coefficient).
Table 3 gives the valuesof the areas under the ROC curves for all SLP parameters and the RNFL photographicscores. For the SLP parameters, ROC curve areas ranged from 0.65 to 0.91,whereas for the RNFL photographic scoring system, the ROC curve areas were0.80, 0.84, and 0.84 for the superior, inferior, and global RNFL scores, respectively.The ROC curve area for the best SLP parameter, NFI, was significantly higherthan the ROC curve area for the global RNFL score (P =.03) (Figure 2). Table 3 also shows the sensitivities at fixed specificities (≥80%and ≥95%) for the SLP parameters and the RNFL photographic scores. With95% specificity (cutoff point of 30), the NFI had a sensitivity of 71% fordetection of glaucoma. At the same specificity (cutoff point of 8), the RNFLglobal score had a lower sensitivity (36%). With specificity at 80%, the NFIand RNFL global score had similar sensitivities (88% and 81%, respectively). Figure 3 shows the balance of sensitivityand specificity for each cutoff point for the SLP parameter NFI.
Using 30 as a cutoff point, the NFI classified as having abnormal eyes9 (28%) of the 32 patients suspected of having glaucoma. Six (67%) of these9 patients had glaucomatous-appearing optic discs, whereas 3 (33%) had ocularhypertension and normal optic discs. Using 8 as a cutoff point, the RNFL globalscore classified only 1 (3%) of the 32 patients suspected of having glaucomaas having abnormal eyes. This patient had a glaucomatous-appearing optic discand was also diagnosed as having abnormal eyes by the NFI.
For both the superior and inferior hemifields, as well as for globalRNFL measures, we found moderate correlations between the amount of RNFL damageas assessed by RNFL photographs and GDx VCC software–provided parameters.Because semiquantitative RNFL photographic assessment has been shown to precededetectable optic disc damage and visual field defects in glaucoma,1,3 the measurements of RNFL thicknessobtained with the GDx VCC may provide clinically relevant information fordiagnosing and monitoring glaucoma patients and also for detecting early glaucomatousdamage.
The correlations found in our study between the SLP parameters and theNiessen semiquantitative RNFL scoring system ranged from a minimum R2 of 15% to a maximum R2 of 47%, depending on the specific SLP parameter evaluated. These valueswere generally better than previously reported correlations in studies comparingthe same RNFL photographic scoring system with earlier versions of SLP usingfixed corneal compensation.12,13 Niessenet al12 found a maximum R2 of 28% for the correlation betweencross-section RNFL measurements obtained with a nerve fiber analyzer (GDxNerve Fiber Analyzer; Laser Diagnostic Technologies Inc) and the semiquantitativeRNFL photographic scoring system. In another study, Zangwill et al13 found R2 valuesranging from 0.35% to 25.7% for the correlation between GDx Nerve Fiber Analyzer(Laser Diagnostic Technologies Inc) parameters and the Niessen photographicscoring method. These findings agree with recent reports that showed an improvementin the correlation coefficients for associations between SLP parameters andother measures of glaucomatous damage, such as visual field indices or opticalcoherence tomography RNFL thickness measurements, when VCC rather than fixedcorneal compensation is used.34,35
The improvement in correlation coefficients with the use of SLP withVCC seems to be most noticeable for thickness parameters than for ratio andmodulation ones. In the study by Zangwill et al,13 whichused SLP with fixed corneal compensation, the best correlation using a thicknessparameter was found between the superior average parameter and the superiorhemifield RNFL score. However, the R2 forthis correlation was only 5%, whereas in our study the same association hadan R2 of 25%. For the ratio and modulationparameters, the best correlation in the study by Zangwill et al13 wasfound for the ellipse modulation parameter and global RNFL score, with an R2 of 25.7%. In our study, the same associationhad a comparable R2 of 21.7%. The strongerimprovement found for thickness parameters than for ratio and modulation parametersis likely because the latter may already compensate for some of the changesin retardation measurements caused by an inadequate corneal compensation insome patients. This finding is supported by the study by Weinreb et al,26 which showed that the diagnostic ability of severalSLP parameters at classifying eyes as glaucomatous or healthy is improvedconsiderably with SLP using VCC as compared with SLP using fixed corneal compensation,particularly for the thickness parameters.
The best correlation in our study was found between the GDx VCC software–providedparameter NFI and the global RNFL photographic score, with an R2 of 47%. This is not a surprising result, because theNFI is calculated using a sophisticated machine learning classifier methodthat takes into account several parameters of the RNFL and is intended toprovide the best measure of the current RNFL status obtained with SLP. Highervalues of NFI were also significantly associated with more severe visual fielddamage, as indicated by the global indices MD and PSD. The correlation betweenNFI and visual field indices was higher than the association found betweenthe visual field indices and the global RNFL photographic score. Interestingly,the correlation between the 2 methods of RNFL assessment was higher than thecorrelation between each method and the visual field indices. This is notsurprising, since RNFL damage often can precede visual field loss1 and therefore structure-function correlations areexpected to be weak at a given point in time.
We also evaluated the ability of the SLP parameters and RNFL photographicscores to discriminate glaucoma patients from healthy individuals. The bestSLP parameter, NFI, had a significantly better diagnostic power than any ofthe RNFL scores evaluated. This was particularly evident for higher specificities.When the specificity was set at 95%, the sensitivity of the NFI to diagnoseglaucoma was 71% compared with only 36% for the global RNFL score. At a moderatespecificity (80%), the NFI parameter and the RNFL global score had similarsensitivities (88% and 81%, respectively). In a study comparing the abilityof several diagnostic tests to distinguish healthy from glaucomatous eyes,Paczka et al36 found that the SLP with fixedcorneal compensation had a similar sensitivity compared with a semiquantitativeRNFL scoring system, when both methods were evaluated at high-specificitysettings.
Besides the apparent better diagnostic ability of the SLP using VCCas compared with the RNFL photographic assessment, other potential advantagesof the SLP technology are notable. SLP provides real-time, immediate, andobjective assessment of the RNFL and does not require pupil dilation.8,37 On the other hand, good-quality red-freeRNFL photographs are technically difficult to obtain, requiring maximum pupillarydilation and clear ocular media.4 In addition,evaluation of RNFL photographs requires special training and remains a subjectivemethod, even when semiquantitative scoring systems are used. However, potentialbenefits of RNFL photographic assessment exist, namely its lower cost andthe fact that this method has been previously validated by longitudinal studiesfor early detection of glaucomatous damage.1- 3
Both the SLP parameters and the RNFL photographic scoring method wereable to find significant differences between glaucoma patients and healthysubjects, as well as between glaucoma patients and patients suspected of havingglaucoma. However, only the SLP was able to detect significant differencesbetween those suspected of having glaucoma and healthy individuals, with 3of the investigated parameters showing statistically significant differencesbetween these 2 groups. Accordingly, using a cutoff with 95% specificity,28% of those suspected of having glaucoma were classified as having abnormaleyes using the NFI parameter, whereas only 3% of the same subjects were classifiedas having abnormal eyes using the RNFL global photographic score. If the excessof abnormal eyes in those suspected of having glaucoma over healthy subjectsis taken as a measure of true glaucomatous defect among the group with suspectedglaucoma, a larger proportion of eyes with early glaucoma damage were identifiedwith the NFI compared with the RNFL photographic scoring method. Althoughthis might indicate an advantage of SLP over conventional RNFL photographicassessment for the identification of early damage in those suspected of havingglaucoma, longitudinal studies are still needed to verify this possibility.
In conclusion, a good correlation was found between RNFL assessmentusing red-free photographs and selected SLP parameters, suggesting that these2 methods provide comparable information about the status of the RNFL damagein glaucoma. In high-specificity settings, the best parameter from the SLPhad better performance than RNFL semiquantitative photographic assessmentin the detection of patients with glaucomatous visual field defects and alsoin the identification of abnormalities in the eyes of patients suspected ofhaving glaucoma.
Corresponding author: Robert N. Weinreb, MD, Hamilton Glaucoma Center,University of California, San Diego, 9500 Gilman Dr, La Jolla, CA 92093-0946.
Submitted for publication April 24, 2003; final revision received September7, 2003; accepted October 1, 2003.
This study was supported in part by the Foundation for Eye Research(Dr Medeiros) and National Eye Institute grant EY11008 (Dr Zangwill).