Technologies for Analyzing the Optic Disc, RNFL, and Macula
Lindsey S. Folio, BSc; Gadi Wollstein, MD; andJoel S. Schuman, MD
Glaucoma can be characterized as a slowly progressingoptic neuropathy that damages the retinalganglion cells (RGCs) and their axons. Thisdamage causes thinning of the retinal nerve fiber layer(RNFL), leads to cupping of the optic nerve head (ONH),and usually results in observable visual field loss. Becauseglaucomatous damage is irreversible, early detection iscrucial for effective treatment through medical or surgicalmethods to halt further progression and preserve functionalvision. Studies have shown that up to half of theRGCs could be damaged before defects are reported onthe visual field.1 A more resolute, objective method ofidentifying glaucoma is therefore needed for at-riskpatients.
Ophthalmic imaging techniques such as optical coherencetomography (OCT) allow objective, quantitativeevaluation of ocular structures. OCT has been shown to correlate well with histological retinal measurements2and to allow direct visualization and quantification of thestructures of the retina and ONH. This device is a valuabletool for detecting glaucoma3-5 and monitoring itsprogression.6
Time Domain OCT
OCT was first described by Huang et al in 1991 as ahigh-resolution, cross-sectional imaging technique thatfunctions similarly to ultrasound but uses light instead ofsound.7 The imaging process involves low-coherence nearinfraredlight, which is split and directed at a moving referencemirror and the tissue of interest. The two lightbeams are then reflected and recombine at an interferometer,where an interference pattern is produced andassessed by a photodetector. This procedure is repeatedalong a line to form a cross-section of the tissue imaged.Time domain OCT (TD-OCT) has an axial resolution ofapproximately 10 µm and can scan at an average rate of400 axial scans/sec.
Spectral Domain OCT
Spectral domain OCT (SD-OCT) for ophthalmic imagingwas first described by Wojtkowski et al in 2002.8 UnlikeTD-OCT, SD-OCT does not require the mirror in the referencearm to move, because all information is encodedsimultaneously in the frequency spectrum. SD-OCT allowsscanning speeds of up to 55,000 axial scans/sec with axialresolutions of 5 to 6 µm.9 The faster scanning speed permitsthe acquisition of three-dimensional datasets, whichallows extensive postprocessing of the data without requiringthe patient to remain in front of the device. Theincreased resolution has resulted in improved visualizationof retinal layers. Some studies have shown SD-OCT imagingto have a greater diagnostic ability for glaucoma comparedwith TD-OCT,10,11 whereas others have shown equivalence.12,13 Table 1 shows currently commercially availableSD-OCT devices.
IMAGING STRUCTURES FOR GLAUCOMA
The common scanning patterns used to assess theoptic disc in glaucoma include a high-density raster scan(cube) or a radial scan. Figure 1 was obtained with anOptic Disc Cube 200 X 200 scan (Cirrus HD-OCT; CarlZeiss Meditec, Inc., Dublin, CA) of a patient with glaucomathat covered a 6- X 6-mm area centered on the ONH.The OCT image is analyzed using automated detection ofthe disc’s boundaries. Cup-to-disc ratios, disc and rimarea, and cup-volume measurements are calculated fromthe data and displayed. Images can be corrected for ocularmovement during scanning by aligning the OCT datausing the en face image (Figure 1A). The en face imagerepresents an aerial view of the scanned area and can beused to detect ocular movement during image acquisitionthat might affect the reliability of the reported measurements.Eye movements will cause discontinuities in bloodvessels and an anomalous appearance to the disc. Studieshave shown that analysis of the ONH with SD-OCT candifferentiate between healthy and glaucomatous eyes.14
Peripapillary thinning of the RNFL is a hallmark ofglaucoma and represents axonal loss due to the RGCdamage. In OCT imaging, the RNFL appears as a highlyreflective layer at the vitreoretinal interface that is automaticallysegmented by the software. An RNFL thicknessmap (Figure 1B) is generated from the scanned cube,allowing visualization of potential RNFL thinning withinthe cube. Circumpapillary analysis can be performed bysampling from the cube of data along a 3.4-mmdiametercircular scan centered on the ONH (Figure 1Aand C). Alternatively, RNFL thickness can be obtained by performing a 3.4-mm-diameter circular scan (rather thana cube), but sampling from a cube scan allows the locationof the circle to be repositioned and matched betweenvisits.
RNFL thickness is presented with a comparison to anormative distribution of age-matched healthy subjects.Typically, the RNFL thickness profile demonstrates anincreased thickness in the superior and inferior positions,giving a double-hump pattern. OCT image analysissoftware also reports the average RNFL thickness as wellas quadrant and clock-hour thicknesses in the 3.4-mmcircle. Measured quadrant and clock-hour thicknessescan also be compared to a normative distribution ofage-matched healthy subjects and are important in testingfor local glaucomatous defects. OCT has been shownto produce accurate and reproducible RNFL thicknessmeasurements.15 Additionally, some studies have shownSD-OCT analysis of the RNFL to be a better diagnostictool for glaucoma than analysis of the ONH.16
Glaucoma damages the RGCs, and approximately 50%of these cells are located in the macular region.17 Theyhave been observed to compose 30% to 35% of the retinalthickness in the macula, making the macular inner layersa potential target for glaucoma detection.18 Severalscanning patterns are commonly used to image the macula,including a high-density raster scan (cube), spoke patternradial scans centered on the fovea, and a mesh scanpattern (combination of vertical and horizontal scans).With the improved resolution and higher sampling rate ofSD-OCT, automated quantification of the inner retinallayers is possible, thus specifically targeting the layersprone to glaucomatous damage. The ganglion cell complex(GCC) analysis (Figure 2) has been shown to providea better glaucoma diagnostic parameter than total retinalthickness of the macula.19
MONITORING GLAUCOMATOUSPROGRESSION WITH OCT
Glaucomatous progression can be identified in theform of narrowing of the neuroretinal rim and thinningof the RNFL and/or GCC. The proven reproducibility ofRNFL measurements obtained with OCT designate thesevalues as potential parameters to measure structuralchanges over time.15 Both TD-OCT and SD-OCT usesoftware that assesses progression by plotting a linearregression of RNFL thickness against age (Figure 3).Progression studies have found TD-OCT progressionanalysis capable of identifying local and diffuse loss ofRNFL, and they have shown the rate of RNFL thinning toexceed the rate of progression determined by visual fieldtesting.6,20
Because SD-OCT is a relatively recent technology,more longitudinal data need to be collected beforestudies can test its ability to detect disease progression.We predict SD-OCT will surpass TD-OCT’s abilitybecause of the higher sampling densities, improvedscanning density, better reproducibility,21 and its imageregistrationcapabilities.
Lindsey S. Folio, BSc, is a research specialist for theDepartment of Ophthalmology at the University ofPittsburgh School of Medicine and a graduate student inthe Department of Bioengineering, Swanson School ofEngineering at the University of Pittsburgh. She acknowledgedno financial interest in the products or companiesmentioned herein.
Gadi Wollstein, MD, is an associate professorof ophthalmology at the University of PittsburghSchool of Medicine and the director of theOphthalmic Imaging Research Laboratories atthe UPMC Eye Center. He has received researchfunds from Carl Zeiss Meditec, Inc., and Optovue Inc.Dr. Wollstein may be reached at email@example.com.
Joel S. Schuman, MD, is the Eye and EarFoundation professor and chairman of theDepartment of Ophthalmology at the Universityof Pittsburgh School of Medicine, and he is thedirector of the UPMC Eye Center. He is also aprofessor of bioengineering at the University of PittsburghSchool of Engineering and a professor at the Center for theNeural Basis of Cognition, Carnegie Mellon University andUniversity of Pittsburgh. Dr. Schuman has received lecturefees and payment of faculty travel expenses from Pfizer, Inc.He receives royalties from intellectual property licensed byM.I.T. to Carl Zeiss Meditec, Inc.
Neil T. Choplin, MD
Scanning laser polarimetry (SLP) uses a physical propertyinherent to the retinal nerve fiber layer (RNFL) toassess its thickness in vivo with a high degree of sensitivityand specificity for diagnosing glaucoma.
Of the three computerized scanning imaging techniquescurrently available to evaluate the RNFL, SLP is theonly one that uses a physical property of the RNFL otherthan reflectivity to make its assessments. This property,called form birefringence, arises in a material or tissue composedof substructures smaller in diameter than the wavelengthof light used to image it. A polarized light passingthrough such a tissue will undergo a measurable phaseshift, called retardation, which is directly proportional tothe thickness of the tissue. Microtubules contained withinthe individual fibers of the RNFL give rise to its birefringentproperties.1
The commercially available scanning laser polarimeter,the GDx VCC (Carl Zeiss Meditec, Inc., Dublin, CA) assessesthe RNFL by passing a polarized light through it andmeasuring the resultant retardation by an ellipsometer.Although retardation is measured in angular degrees, it isproportional to thickness and is expressed in microns. This is based on the relationship betweenthe amount of retardation and the histologicallydetermined RNFL thickness foundin monkey eyes.2 Birefringence inherent tostructures in the anterior segment, mostlythe cornea, is subtracted by a compensatorthat is individually adjusted for theeye being assessed (variable corneal compensation[VCC]).3 The latest version ofthe instrument incorporates a softwarecorrection for eyes with low signal-to-noiseratios called enhanced corneal compensationor ECC. Using individualized anteriorsegment compensation allows the resultsof SLP to better match the appearance ofthe RNFL by red-free fundus photography.4
After patient-identifying information, including birthdate and race, are entered, SLP assesses anterior segmentbirefringence with the method described by Zhou andWeinreb.3 The software automatically adjusts the anteriorsegment compensator. RNFL thickness can then beassessed with eye-specific corneal compensation in a 20°X 20° field of view at a resolution of 128 pixels X 128 pixels.Usually, the right eye is measured first, followed by theleft eye.
The GDx software positions a circle, 8 pixels wide withan inner diameter of 54 pixels, so that it is centered on theoptic nerve image. This circle is known as the TSNIT plot,because it contains RNFL values going from the temporalside of the optic nerve, then superiorly, nasally, inferiorly,and back to temporally. Based on the retardation valueswithin this band, the software calculates six parameters:TSNIT average, superior average, inferior average, TSNITstandard deviation, inter-eye symmetry, and nerve fiberindicator (NFI). The NFI is the value derived from a supportvector-machine—derived algorithm trained to discriminatebetween healthy and glaucomatous eyes.Possible values range from 1 (normal) to 100 (glaucoma)on a linear scale.
Results are displayed for both eyes on a single page(Figure 1). The printout features a reflectance image thatis used for orientation and assessment of image quality.The retardation image (thickness map) presents themeasurements of retardation in a 20° X 20° area aroundthe optic nerve head, color coded for value. Bright, warmcolors represent thicker areas, and dark, cool colors representthinner areas. Parameters and the TSNIT plot areconstructed from measurements within the measurementcircle. The deviation map shows how the measurementscompare to an age- and race-matched normative database.Points that are outside normal limits are flaggedaccording to statistical significance. This map is comparableto the Humphrey Field Analyzer’s (Carl Zeiss Meditic,Inc.) total deviation probability plot. The TSNIT plot(temporal-superior-nasal-inferior-temporal) displays retardationin the measurement circle (dark line). The normal95% range is shown as a shaded area; the yellow area representsvalues between the first and fifth percentiles,whereas values in the red area are below the first percentile.The parameters that are within normal limits(above the fifth percentile) are shown in green on theprintout. Parameters that are outside normal limits areshown in white on a shaded background, with the shadingcorresponding to its percentile in the normative database(yellow between the first and fifth percentile, redbelow the first percentile).
The NFI is the one parameter of the GDx VCC showingthe greatest ability to discriminate between glaucomatousand normal eyes.5 In a group of 73 healthy subjects and146 glaucoma patients of similar age, the sensitivity andspecificity of the NFI were 89% and 96%, respectively, at acutoff value of 40 (GDx VCC software version 5.0.1). Atthe same specificity of 96%, the sensitivity to detect mild,moderate, and severe glaucoma—classified by the severityof the patients’ visual field mean deviation score—was84%, 87%, and 100%, respectively. The relatively low sensitivityfor early glaucoma may be due to the fact that theparameters are relatively insensitive to focal defects; thesewould easily be seen on the deviation map. With the latestsoftware versions of the GDx VCC (version 5.3.1 andlater), similar results are seen with the cutoff level at 35.
INTERPRETATION OF PRINTOUTS
When interpreting GDx VCC measurements, cliniciansshould expect some variability in the appearance of the RNFL. In addition, they should keep in mind that one outof every 20 healthy subjects might be expected to haveparameters that are flagged at P < 5%. In the retardationmaps of healthy eyes, retardation is always present adjacentto the thicker blood vessels superior and inferior tothe optic disc. The appearance of the RNFL may vary significantlybetween subjects. The deviation map usuallyshows no flagged pixels. Flagged (false positive) areas mayoccur in the nasal half of the map in healthy subjects. TheTSNIT plot shows a “double-hump” pattern (correspondingto a thicker RNFL superiorly and inferiorly), althoughthe appearance of the TSNIT plot may vary considerablybetween subjects. The plot is usually within the greenarea. In healthy eyes, the TSNIT plots of both eyes aresymmetrical. Asymmetry may occur in the nasal areas. Inhealthy eyes, the NFI is usually below 35.
In glaucomatous eyes, defects of the RNFL oftendevelop superotemporal and inferotemporal to theoptic disc. Loss may be localized, is sometimes visible asa clear wedge-shaped defect, and may be diffuse. In theretardation maps of glaucomatous eyes, retardation islost adjacent to the thicker blood vessels, especiallysuperotemporal and inferotemporal to the optic disc.The deviation maps show areas of flagged pixels (withP < .5% and P < 1%) superotemporal and inferotemporalto the optic disc. The TSNIT plots are often belowthe normal range in the temporal part of the superiorand inferior bundles in mild-to-moderate glaucoma. Inaddition, the TSNIT plots in the superotemporal andinferotemporal regions are asymmetrical between eyes.Asymmetries in the nasal sectors are less specific, becausethey occur in healthy subjects as well. For eyeswith severe glaucomatous damage, the TSNIT plot maybe flat. The NFI is usually 35 or higher, but some localizeddefects may not be picked up by this parameter.
DETECTION OF PROGRESSION
Software has recently been introduced to help detectsignificant RNFL changes compatible with a progressiveloss of tissue. Called Guided Progression Analysis (GPA), itis analogous to the tool developed for the analysis of visualfield progression. This statistical tool compares theinterexamination variability of the individual patient tothat observed in a matched population. Reproduciblechanges in the RNFL can be identified suggestive of glaucomatousprogression.
SLP has proven to be a useful clinical tool for the assessmentand follow-up of eyes with glaucoma and otheroptic neuropathies. Detecting RNFL loss prior to visualfield loss can help the clinican decide to institute therapyearlier, whereas establishing the absence of loss in an eyesuspected of being glaucomatous can prevent unnecessarytreatment.
Neil T. Choplin, MD, is a glaucoma specialistwith Eye Care of San Diego in California. Heacknowledged no financial interest in the productsor company mentioned herein. Dr. Choplinmay be reached at firstname.lastname@example.org.
Confocal ScanningLaser Tomography
Paul H. Artes, PhD
One of the most rewarding things about my involvementin glaucoma research during the past fewyears has been to witness the huge changesbrought about by technological advances in ocular imaging.The Heidelberg Retina Tomograph (HRT; HeidelbergEngineering GmbH, Heidelberg, Germany) was among thefirst of such devices, and it is now one of the most usefuland reliable workhorses in ophthalmic practice. The instrumentis in its third generation, and the company’s commitmentto the platform’s stability has ensured that the dataare largely backwards compatible. This means that manycenters now have long data series from individual patientswith glaucoma, which, in turn, helps to establish and refinetechniques for looking at change.
This article reviews some of the practical issues of usingthe HRT to document the status of the optic nerve head(ONH) and the retinal nerve fiber layer (RNFL) in glaucomaas well as to look for progression.,/p>
The HRT3 provides two analyses to help interpret a singleimage for the likelihood of damage to the optic disc. Theclassic approach, the Moorfields Regression Analysis (MRA),compares the neuroretinal rim area of the whole disc, and in six sectors, to the size of the ONH.1 The process requires theuser to place a contour line along the edge of the optic disc,a fairly easy task that can usually be done within seconds bymarking just four points along the cardinal positions.
Another tab in the Eye Explorer software reveals the glaucomaprobability score (GPS), a shape analysis of the ONHthat does not require a contour line.2 This option is helpfulmostly to inexperienced users who may not yet feel confidentabout outlining the disc. In most cases, the diagnosticperformance of the GPS is similar to that of the MRA.3Stereometric parameters derived from the contour line,however, provide quantitative information on the opticdisc’s size and neuroretinal rim area that can later be used asa comparison with subsequent images.
Both the MRA and GPS tend to “overcall” abnormality inlarge healthy discs. Caution is also needed with very smallONHs in which a loss of rim area does not become obviousuntil late in the disease.3,4 A “quantile regression” approachfor the MRA that improves the specificity in large discs wasrecently published and will, I hope, soon be incorporatedinto the commercial software.5
The HRT does not directly measure the thickness of theRNFL but derives its height profile around the contour line.The familiar double-hump pattern can be a valuable diagnosticsign. One subjective technique that I find particularlyuseful is to pay attention to the fine striations of the RNFLaround the ONH. Owing to the excellent optics and theconfocal nature of the scanning system, localized defects inthe RNFL often appear much more obvious in the HRTreflectance image than on clinical examination. The “movie”feature of the software provides a slow-motion view alongthe depth axis through the confocal optical sections, andthis can also give a superb view of wedge-type RNFL defects.
In a glaucoma practice, perhaps the greatest utility ofthe HRT lies in its ability to help the clinician assesschanges in the optic disc over time. In principle, there aretwo different approaches for this function. The first is anevent-type analysis, which focuses on the differencebetween the current image and an earlier baseline. Thesecond is a trend-type approach, which derives the rate ofchange over time in one of the stereometric indices suchas rim area. In clinical practice, event-type analyses arearguably more relevant, because physicians must makedecisions after short intervals. Trend analyses tend to bemore useful with long follow-up (at least 4 years withimaging every 6 months).
The topographical change analysis is an event-type analysisthat is a part of the Eye Explorer software of the HRT.This analysis compares the surface height measurements offollow-up images to those of the baseline image. Changes that are consistent (ie, those that are present in three out offour follow-up images) are highlighted in a color-codedoverlay on top of the reflectance image. By inspecting thesemaps alongside the reflectance and topographic images,clinicians can determine whether and where a change islikely to have taken place (Figure 1).
Of the many stereometric parameters, rim area appearsto be the intuitive choice for measuring the speed ofchange over time. Because the contour line drawn in thebaseline image is automatically aligned to the follow-upexaminations, an objective rate of change can be derivedfrom a series of measurements over time (Figure 1). Dataseries with substantial follow-up are now available fromseveral centers, and it is likely that an evidence-based andvalidated clinical tool for estimating rates of change will bemade available within Heidelberg’s Eye Explorer software inthe near future.
A FEW POINTERS
It is often a good idea to obtain two or even threeimages during a single session. Once the patient is set upin front of the instrument, the additional imaging onlyadds a few seconds to the procedure. The quality of thebaseline image is critical for subsequent analyses of progression,so selecting the best of several images can greatlyincrease the odds of detecting subtle change. Also, anyvariability observed between the images obtained on thefirst sitting will give the clinician a useful picture of whatvariation can be expected subsequently, and this will helphim or her to interpret any data obtained in the future.
Measuring progression in clinical data is complex. Often, physiciansneed an answer, even if only approximate, after just a fewexaminations. This is quite unlike the orthodox approach of testingstatistical hypotheses and much more in the spirit of exploratoryanalysis. A strength of the Eye Explorer software is that it empowersthe user in this regard: it is easy to change a baseline examination orto exclude an image of low quality with just a single click of themouse. By using the software interactively, the clinician can get amuch more comprehensive idea of the evidence—something thatis impossible to do with a single static printout.
Excellent image quality is critical for the analysis of progression,and images with standard deviations greater than 30 µm are unlikelyto be useful for tracking change over time. Moreover, as withother tests, a single “bad apple” can dramatically influence the findingsin a series. Clinicians should review both the reflectance andthe topographic images, even though it may take a little practice toget familiar with interpreting the false-color representation of thetopography.
At present, the HRT is the most thoroughly validated tool forassessing structural progression at the optic disc.6-9 Tools such asthe topographical change analysis can help users to gauge change,but they augment and assist rather than replace clinical judgment.Newer technologies such as spectral-domain optical coherencetomography have the potential to provide more precise measurements,but it will take time before the knowledge base and clinicalevidence for interpretation reaches the maturity of that availablefor the HRT.
Paul H. Artes, PhD, is an associate professor and foundationscholar in glaucoma research at Dalhousie Universityin Halifax, Nova Scotia, Canada. He has received researchsupport from Heidelberg Engineering GmbH and fromOptovue Inc. Dr. Artes may be reached at email@example.com.