A, Comparison of photographic fields.Photograph of right eye of Diabetic Retinopathy Study standard field 2 takenwith 30° film camera. B, Similar standard field 2 taken with the 45°nonmydriatic digital camera. A standard grading grid is centered on the fovea.
Klein R, Meuer SM, Moss SE, Klein BEK, Neider MW, Reinke J. Detection of Age-Related Macular Degeneration Using a NonmydriaticDigital Camera and a Standard Film Fundus Camera. Arch Ophthalmol. 2004;122(11):1642-1646. doi:10.1001/archopht.122.11.1642
To compare gradings of lesions associated with age-related macular degeneration(AMD) from digital and stereoscopic film images.
Instrument validation study.
Sixty-two subjects (124 eyes) with varying degrees of AMD, includingno AMD.
Images of the optic disc and macula were taken using a 45° digitalcamera (6.3 megapixels) through dark-adapted pupils and pharmacologicallydilated pupils. In addition, 30° stereoscopic retinal film images weretaken through pharmacologically dilated pupils of the same eyes. All imageswere graded for drusen size, type, and area; pigmentary abnormalities; geographicatrophy; and neovascular lesions using the modified Wisconsin Age-RelatedMaculopathy Grading System. Exact agreement and unweighted κ scores werecalculated for paired gradings resulting from digital and film images.
Main Outcome Measure
Agreement between gradings obtained from stereoscopic slide transparenciesand digital nonstereoscopic images.
Exact agreement between gradings of digital and stereoscopic film imagestaken through pharmacologically dilated pupils was 91% (κ = 0.85)for the categories of none, early AMD, and late AMD. Exact agreement for gradingsof digital images taken through dark-adapted pupils compared with gradingsof film images was 80% (κ = 0.69). Exact agreement for gradingsof digital images captured through dark-adapted and pharmacologically dilatedpupils was 86% (κ = 0.78). In addition, κ scores for agreementbetween different approaches for individual lesions were moderate to almostperfect.
Gradings resulting from high-resolution digital images, especially whenthe pupil is pharmacologically dilated, are comparable with those resultingfrom film-based images. We conclude that digital imaging of the retina isuseful for epidemiological studies of AMD.
Age-related macular degeneration (AMD) is a major cause of visual impairment.1,2 Fundus photography with film-basedcameras has routinely been used to document the presence and severity of AMDin clinical practice and epidemiological studies.3,4 Optimally,fundus photography is performed through a pharmacologically dilated pupil.However, in situations in which mydriatic agents cannot be used, photographywith film-based nonmydriatic cameras has successfully been used as an alternativemeans of imaging for AMD.5- 7
Recently, new high-resolution digital cameras have become availableand are being used to detect AMD in research and clinical settings.8- 10 The purpose of thisarticle is to compare the severity of AMD using images of the same retinasrecorded with 3 different imaging systems: nonstereoscopic color 45° retinalimages taken with a digital camera through a dark-adapted pupil, nonstereoscopiccolor 45° retinal images taken with a digital camera through a pharmacologicallydilated pupil, and stereoscopic color 30° retinal images taken with astandard film camera through a pharmacologically dilated pupil. In addition,patient acceptance of these imaging systems was compared.
Our study included 62 patients seen at the University of Wisconsin RetinaClinic (Madison). We attempted to recruit at least 30 patients with a diagnosisof AMD. Tenets of the Declaration of Helsinki were followed, and approvalfrom the institutional human experimentation committee was granted. Informedconsent was obtained from each participant. Birth dates and history of cataract,cataract surgery, diabetes, AMD, and retinal photocoagulation were ascertainedin 58 subjects (4 patients did not provide a history).
Participants were seated in a darkened room. Pupil size was estimatedby comparing it to progressively increasing circles from 1 to 9 mm in diameter.Both eyes of each participant were photographed in a similar fashion usinga 45° 6.3-megapixel digital nonmydriatic camera (Canon, Lake Success,NY). This camera used an infrared light to televise a view of the fundus througha dark-adapted pupil. Field location and focusing of the retinal image wereaccomplished by the photographer’s use of a laptop computer. Two photographicfields were taken of each eye, the first centered on the optic disc and thesecond centered on the fovea (Figure).The 45° image was minified ×0.64 (at 0 diopters) compared with thattaken with a standard 30° camera (Zeiss FF4; Carl Zeiss, Inc, Jena, Germany)(Figure). The iris color was determinedby direct observation and was recorded as gray/blue, yellow/green, or tan/brown.The presence of any corneal or lens opacity was recorded. The photographerrecorded the lengths of time necessary for pupil dilation and fundus photography.
One drop of 2.5% phenylephrine hydrochloride and 1 drop of 1% tropicamidewere then instilled in the cul-de-sac of each eye to obtain dilation. Thesize of the dilated pupil was estimated. Two images of the same retinal fields,as described previously, were taken with the nonmydriatic digital camera throughthe pharmacologically dilated pupil.
Stereoscopic retinal photographs were taken with a standard 30°fundus camera (Zeiss FF4) centered on the disc (Diabetic Retinopathy Study11 standard field 1) and macula (standard field 2),and a nonstereoscopic color fundus photograph was taken temporal to but includingthe fovea of each eye.12,13 Thetimes at the beginning and end of photography were recorded. Subjects wereasked to evaluate their comfort with the flash of each camera on a 10-stepscale ranging from 1 (no discomfort) to 10 (extremely uncomfortable) aftereach set of photographs. At the end of the photography session, participantswere asked the following question: “Which is least tolerable: havingyour pupils dilated, the flash, both the same, or both tolerable?”
The digital images were graded using the standard AMD protocol. Graderswere masked with respect to information about the subject. Digital imagesof both eyes were graded by the same grader; the right and left eyes fromthe same photography session of each participant were both displayed to thegrader. Graders were asked to judge field definition, focus of the photographs,pupil size, and the appearance of artifacts prior to determining AMD severitylevel. Each image was graded twice (a preliminary and a detail grade) onlineusing the Multi-Ethnic Study of Atherosclerosis protocol,14 amodification of the Wisconsin Age-Related Maculopathy Grading System.13 Every digitized image was graded twice by the samegrader for the preliminary grading. For the first grading, there was no imagemanipulation except magnification (no contrast enhancement, lightening, orred-free images). For the second grading, the full complement of image enhancementtools was available.
If the preliminary and detail gradings agreed, the grading was consideredfinal. If there was disagreement between gradings for a lesion, the imagewas sent to an edit grader for reevaluation of that lesion without knowingwhat the specific disagreement was. The preliminary, detail, and edit gradingswere compared again for agreement. If the edit grading agreed with eitherthe preliminary or detail grading, that one was considered final. If a disagreementremained, the grading of the image was adjudicated by the Reading Center codirector(R.K.). Nine different graders were assigned to perform either preliminary,detail, or edit gradings for each image. No grader saw the same eye twice.
All stereoscopic photographs taken with the film-based 30° cameraused color 35-mm slide film (Ektachrome 100 Plus Professional; Kodak, Rochester,NY), which was processed and returned as 2 × 2-in slides.The slides were mounted in clear plastic mounting sheets and graded usinga light box and a Donaldson stereo viewer with original magnification ×5.Two gradings for AMD were performed for each eye.12 First,a preliminary masked grading was done by one senior grader for drusen size,type, and area; pigmentary abnormalities; geographic atrophy; and exudativelesions. Next, detailed gradings were performed by other experienced graders.For detailed grading, each eye was graded independently of the fellow eye.The assessment consisted of a subfield-by-subfield, lesion-by-lesion evaluationof each photograph set using the Wisconsin Age-Related Maculopathy GradingSystem.13,15 Then a series ofedits and reviews was performed similar to that done in the digital grading.The presence and severity of specific lesions (eg, maximum drusen size, type,and area and pigmentary abnormalities), as determined by detailed grading,were compared with those of the preliminary grading. Standardized edit ruleswere used to adjudicate disagreements.13,15
Early AMD was defined as the presence of soft indistinct drusen onlyor either hard or soft drusen and pigmentary abnormalities (increased retinalpigment or retinal pigment epithelial [RPE] depigmentation) in the absenceof signs of geographic atrophy or exudative lesions. Late AMD was definedas the presence of either geographic atrophy or signs of exudative maculardegeneration.
Differences between means were tested for statistical significance usingthe t test or, in the case of more than 2 groups,analysis of variance. Agreement between grading methods was evaluated withthe κ statistic.16,17
The 62 participants ranged in age from 30 to 90 years; the median agewas 64 years. Twenty-six (45%) of 58 had a history of cataract, and 10 (17%)of 58 had a history of cataract surgery. Of the 116 irises, 68 (59%) werejudged to be gray/blue, 14 (12%) yellow/green, and 34 (29%) tan/brown. Wephotographed 101 eyes through dark-adapted and pharmacologically dilated pupilsusing the digital camera and through pharmacologically dilated pupils withthe film camera. Twenty-three eyes were not graded because of non-AMD processes(such as pigmentary dystrophies) or photocoagulation scars. There was no significantdifference regarding participants’ ability to tolerate the differenttypes of photography (data not shown).
Table 1 lists comparisons of AMDseverity by grading of manipulated images taken with a nonmydriatic 45°digital camera through both dark-adapted and pharmacologically dilated pupilsas well as stereoscopic film images taken through pharmacologically dilatedpupils. When categorizing AMD as none, early, or late, agreement was betterbetween digital and film images taken through pharmacologically dilated thandark-adapted pupils (P = .03). Withoutpharmacological dilation, more digital than film images were ungradable forAMD severity level. Exact agreement between digital images of dark-adaptedeyes and those taken after pharmacological dilation for AMD severity levelwas 86.1% (κ = 0.78; SE = 0.05) (Table 1).
Exact agreement for specific AMD lesions was high with moderate to almostperfect κ scores (Table 2). Whendisagreements were present for specific AMD lesions, drusen 125 μm indiameter or greater were more likely to be graded as present (10.9% vs 2.0%)and increased retinal pigment (3.0% vs 5.0%), RPE depigmentation (1.0% vs5.0%), and RPE detachment (1.0% vs 5.9%) were less likely to be graded aspresent in manipulated digital images of dark-adapted pupils compared withfilm images of pharmacologically dilated pupils. Drusen 125 μm in diameteror greater were more likely to be graded as present (5.9% vs 2.0%) but RPEdepigmentation (4.0% vs 5.9%) and RPE detachment (2.0% vs 5.9%) were lesslikely to be graded as present in manipulated digital images compared withfilm images of pharmacologically dilated pupils.
The mean ± SD score on the comfort scale varied from2.4 ± 0.8 (1 being the most comfortable and 10 the leastcomfortable) for digital photography through dark-adapted pupils to 2.5 ± 2.0for digital photography and 3.9 ± 2.6 for film-based photographythrough pharmacologically dilated pupils. The differences among these meanswere statistically significant (P<.001). Personswith brown or green eyes were more likely to give higher scores for discomfort(P = .01) than gray- or blue-eyed persons.There was no effect of age or pupil size on comfort with digital or film photography(data not shown).
Our study demonstrated some of the strengths and limitations of boththe nonmydriatic digital 45° camera and the standard 30° film-basedcamera. The advantages of the nonmydriatic digital 45° camera in contrastto the standard 30° film-based camera were that it (1) was less expensive;(2) took a shorter time to learn to use; (3) provided excellent resolutionand images that could be magnified and further manipulated (eg, use of thegreen channel as well as lightening or darkening the image to bring out thepresence of a lesion); (4) provided immediate feedback to the photographerregarding the quality of the photograph and to the participant regarding thepresence of abnormalities; and (5) was not necessary to pharmacologicallydilate the pupil before taking photographs. The relative disadvantages ofthe nonmydriatic digital camera were that (1) the resultant image did nothave stereopsis; (2) there was a relative decrease in color contrast comparedwith film in eyes that had very red fundi; and (3) there was a higher frequencyof ungradable photographs (especially in the presence of small nonpharmacologicallydilated pupils and/or media opacities). Use of the cameras was similar interms of participant acceptance.
Data from our study show moderate to almost perfect agreement betweenthe digital and film-based cameras for detecting AMD and its lesions. Thedisagreements for AMD severity level were largely due to 45° digital imagesthat could not be graded or were of poor quality. The nonstereoscopic digitalimages may have contributed to the grader’s missing retinal abnormalitiessuch as RPE depigmentation that are seen more easily on stereoscopic images.The findings in our study are consistent with another study that comparedgradings of stereoscopic digital images and film-based images taken througha pharmacologically dilated pupil.8 In thatstudy, the κ value for the between-technique agreement for stages of AMDseverity was approximately 0.76. The lower level of agreement in that studycompared with ours may have been because AMD severity levels were based onunedited and unadjudicated gradings of low-resolution (800 × 600pixels) images, whereas our results were based on edited and adjudicated gradingsusing higher-resolution images (3072 × 2048 pixels).
In our study, there was a higher frequency of larger (≥125 μmin diameter) soft drusen and a lower frequency of increased retinal pigmentand RPE depigmentation on nonmydriatic digital images than film images. Thesedifferences were more marked for images taken through dark-adapted than pharmacologicallydilated pupils, and they were not found in another study when gradings ofstereoscopic digital images were compared with those from stereoscopic film-basedimages (R. van Leeuwen, MD, PhD, written communication, January 2004). Wespeculate that a lack of stereoscopic effect may have led to a poorer abilityto distinguish the edges of large drusen or detect subtle RPE depigmentation,especially when the contrast was poor. However, this did not greatly affectthe agreement between gradings using the 2 photography approaches for detectingAMD severity in our study. The decreased ability to detect pigmentary abnormalitieswith digital photography was consistent with findings from a recent studyexamining the agreement between gradings of digitized images that were madefrom film-based images using the original film slides.18 Inthat study, agreement was good for detecting RPE depigmentation but poor fordetecting increased retinal pigment. This might have resulted from a lackof capturing pigment in scanning the film image. It is also possible thatdifferences in spectral sensitivity between film and digital imaging contributeto this discrepancy.
Digital photography provides the grader with powerful tools to examinepoor-quality images. For example, without pharmacological dilation, the smallpupil may lead to relatively dark images that might obscure AMD lesions. Insome cases, manipulating the brightness in such images provides easier detectionof these lesions. We had expected that for AMD lesions, manipulated digitalimages would be more closely correlated with film images than unmanipulateddigital images. However, we did not find this to be the case (R.K., unpublisheddata, 2004). Caution must be exercised when manipulating digital images toavoid the introduction of artifacts that might result in false-positive orfalse-negative results. For this reason, we have developed protocols thatspecify allowable manipulations of magnification, contrast, and brightnessas well as use of specific color channels with the digital images. The findingsfrom the manipulated images must be seen on the original unmanipulated imagesfor the lesions to be graded as present.
The appropriateness of using the digital nonmydriatic camera in epidemiologicalstudies of AMD depends on the objectives of the specific study. When feasible,stereoscopic fundus photography taken with a digital or film camera throughpharmacologically dilated pupils is the preferred approach for detecting AMD.If possible, an ongoing epidemiological study or clinical trial should notchange from film to digital capture of retinal images. However, if AMD isnot the primary endpoint in a new study and pharmacological dilation of thepupils is not feasible, digital photography using a nonmydriatic camera shouldprovide an alternative to stereoscopic fundus film photography. Furthermore,the nonmydriatic digital camera offers a distinct advantage compared withthe nonmydriatic film camera because it enables the photographer to take anotherimage of the retina when the first image is poor or ungradable owing to blinkingor other artifacts. Because of its lower flash intensity, the nonmydriaticdigital camera allows photography of 2 or more fields of both eyes in a relativelyshort time compared with the nonmydriatic film camera.
In summary, data from our study show moderate to almost perfect agreementbetween gradings of AMD severity from digital and film-based images. The useof digital photography in epidemiological studies has resulted in the developmentof new protocols and software for the capture, transmission, and reading ofdigital images. New software has enabled application of the Wisconsin Age-RelatedMaculopathy Grading System currently used in grading film images. Furtherstandardization and development of guidelines for using, taking, and gradingdigital images are necessary to allow comparisons of results among differentsystems. This will be especially important in multicenter studies using differentdigital cameras and software to capture images.
Correspondence: Ronald Klein, MD, MPH, Departmentof Ophthalmology and Visual Sciences, University of Wisconsin–Madison,610 N Walnut St, 460 WARF, Madison, WI 53726 (firstname.lastname@example.org). Reprints are not available from the author.
Submitted for publication: March 18, 2004;final revision received July 15, 2004; accepted July 26, 2004.
Financial Disclosure: None.
Funding/Support: This study was supported bygrants EY 06594 and HL 69979 (Drs R. Klein and B. E. K. Klein) from the NationalInstitutes of Health, Bethesda, Md.
Acknowledgment: We thank Canon USA Inc (LakeSuccess, NY) for providing the digital CR6-45NM nonmydriatic retinal cameraand the University of Wisconsin Department of Ophthalmology and Visual SciencesRetina Service for their help in identifying participants in the study.