Study results show substantial agreement between monoscopic digital photographs and stereo 35-mm film in assessing ETDRS severity level. Grading agreement using monoscopic digital photographs also compared well with stereo digital images. The results showed that grading agreement of stereo digital images was comparable to that with stereo film (
Table 2). Given the declining use of film photography in ophthalmology, we did not compare monoscopic film with stereo film. However, our analyses showed high agreement between monoscopic digital images and both the film and digital stereo formats, implying that the grading agreement between monoscopic and stereo slide photography would be similar to our monoscopic versus stereo digital image results. Disagreements in results appear to reflect typical variability in the grading process and fundus photography illumination and focus, rather than intrinsic differences between stereo film, stereo digital, or monoscopic digital photography.
We also found reproducibility of classification of diabetic retinopathy severity in monoscopic or stereo digital images comparable to that with stereo film. Reproducibility compared favorably to that described in ETDRS report 12 (
Table 3).
To the best of our knowledge, this is the only comparative study to apply the ETDRS diabetic retinopathy grading photography protocol and severity classification system to both monoscopic and stereo images. Except for replacing a 35-mm camera's film sensor with a digital chip, we minimized changes to the ETDRS protocol for monoscopic grading. Photographs were taken with the same camera optics, field magnification, area, and size. Digital photographs were balanced for color and contrast to closely match 35-mm fundus slides.
16 Our 2400 × 2000-pixel digital photography resolving power was approximately 13 μm on the retina, sufficient for detecting the smallest IRMA, NVE, or Ma seen on the film. As much as possible, we minimized differences between viewing stereo and monoscopic images. Viewing magnification was similar between stereo film (12.5×) and digital images (13×). For grading monoscopic digital images, each field was displayed full screen at 13×. When grading stereo digital images, each field was first reviewed monoscopically at full screen. Each stereo pair was reviewed at 6.5× to maximally fit both the left and right images to the monitor. Stereo viewing is not only influenced by how images are acquired, but also by how they are displayed.
21 We used stereo viewers that maintain color fidelity and image contrast, avoiding anaglyph (red/blue-green) glasses, polarized viewers, or interlaced liquid crystal shuttering systems. Display monitors were calibrated to ensure adequate and consistent brightness. Readers with a minimum of 10 years of ETDRS protocol grading experience reviewed all images.
Maintaining consistently high quality in stereo fundus photographs is difficult. Dark fundus pigmentation and patient flash tolerance are barriers to maximizing depth of field. Poor focus also limits stereopsis and identification of subtle neovascularization, IRMA, and fibrous proliferation. An experienced reader not involved in grading reviewed the stereo effect in 10% of digital and film image sets. Good stereo effect was found in both media. The stereo effect in sampled photographs was deemed typical for multicenter clinical trials or epidemiologic studies using certified photographers. Some eyes had better stereo effect than others, regardless of medium. Greater or lesser stereo effect seemed to be associated with individual eyes rather than medium.
Readers appeared to be more comfortable in assessing film than digital images, perhaps due to their many years of experience in grading from 35-mm slides. The graders rated their confidence as high in 41 (26.9%) of 152 eyes imaged on film, 16 (10.5%) of 152 eyes with stereo digital images, and 18 (12.1%) of 149 eyes with monoscopic digital images. However, lower levels of confidence did not lead to lower reproducibility among graders using digital stereo or monoscopic images. All graders agreed on the same severity level in 59 (38.8%) of 152 eyes using film, in 63 (41.4%) of 152 eyes using stereo digital images, and in 65 (43.6%) of 149 using monoscopic digital images.
Previous studies have compared monoscopic 35-mm slides
22,23 or monoscopic digital images
24–26 to the ETDRS stereo photography protocol. Moller et al.
22 compared 60°, monoscopic, single-field, 35-mm color slides to the ETDRS 30°, stereo, seven-field, 35-mm color slide protocol. Neovascularization falling within the field area of both protocols was missed in two eyes in grading the 60° images. Low magnification of 60° photography was believed to be the reason.
22 The authors suggested that a higher magnification would be better for achieving depth perception than are wider fields. EURODIAB compared a 45°, monoscopic, two-field photography protocol to the ETDRS reference standard.
27 They cited lack of stereopsis and/or low image magnification as reasons for some grading disagreement. In both studies, the reliability of monoscopic fundus photography was confounded by magnification differences between studied systems and the ETDRS protocol. Most studies have reported pooled ETDRS severity levels rather than the full scale. In some studies, the digital image resolutions were insufficient for detecting small diabetic retinopathy lesions. The lack of comparability across these protocols precludes a systematic literature review of monoscopic digital photography versus the ETDRS stereo film grading protocol.
The interpretation of the results of this study is not without caveats. We selected patients with relatively clear media, a variable that cannot be readily controlled in the course of typical research or clinical care. Some degree of lens opacity is common, as cataracts develop earlier in diabetic patients than the general population. Drug treatments may also have cataractogenic side effects. Our results, therefore, may not be applicable to general diabetic retinopathy patient populations.
It is widely believed that viewing stereo images is critical for distinguishing IRMA abnormalities from extraretinal NVE. Both abnormalities are important index lesions. The absence or presence of IRMA separates level 35 (moderate NPDR) from more severe levels, and the absence or presence of NVE distinguishes NPDR (<level 61) from PDR (≥level 61). Our results show reasonably good agreement between monoscopic digital images versus stereo film in detecting IRMA and NVE (
Table 5). This agreement may be related to the readers' experience in recognizing diabetic retinopathy features and characteristic anatomic sites that are not dependent on stereopsis, such as IRMA and neovascularization, which are common in areas of vascular dropout.
1 Spoke-and-wheel networks,
28 saccular or fusiform dilated tips are also retinal neovascularization features.
6 Neovascularization in areas other than the disc is commonly located at arteriovenous crossings.
29 Large nets of neovascularization can cross over retinal vessels, a telltale sign not dependent on stereopsis
1 and is more frequently found on temporal than nasal veins.
29 The superotemporal quadrant is the most frequent initial site.
30 Most initial neovascularization is within 6 disc diameters of the optic disc. Three-fourths are in fields 1 (optic nerve), 4 (superotemporal), and 7 (inferonasal).
30 Opaque strands or sheets of fibrous proliferation are distinguishable from white, feathery, soft exudates.
6 Preretinal hemorrhages are commonly associated with a fluid level in addition to obscuring retinal vessels.
6
We had a small sample of eyes with NVD and FPD. The eyes with NVD that was missed in monoscopic digital images were assigned appropriate retinopathy severity levels because there were co-existing proliferative index lesions.
Agreement between monoscopic digital images and stereo film grading proliferative eyes was high. We learned from additional review of NVD monoscopic data that most would be detected if the optic disc single image was in good focus, color balance, and proper exposure. The depth perception revealed in a stereo pair may have been helpful in only one NVD eye.
Monoscopic images in this study were not acquired as single images. Instead, we used the left or right image of a stereo pair that exhibited the best focus and contrast. The monoscopic view thus could never be better than its stereo view. This limitation accounted for three ungradable eyes among the monoscopic photographs. Suboptimal image quality of the monoscopic images also accounted for missed NVD. Some stereo views were probably better than their corresponding single image because human vision ignores artifacts in a single photograph when fusing a stereo pair. To achieve adequate stereo base, we took the first photograph of a stereo pair as far to one side of a pupil as possible without introducing shading or edge artifacts. It is likely that our monoscopic image quality would have been better without this requirement. The combined information from both images of a stereo pair compensated for the suboptimal quality of the constituent single image due to the readers' cognitively combining good and poor images.
There are trade-offs in either monoscopic or stereo photography. Sequential stereo imaging creates depth perception but constrains photography to oblique views near the left and right sides of a dilated pupil. This limitation can compromise image clarity in both members of a stereo pair in eyes with suboptimal dilation or substantial peripheral media opacities. Monoscopic imaging allows a straight shot of the fundus, particularly useful in eyes with restricted pupillary dilation or media opacities. However, there is no stereo effect and the reader does not have the insurance of two images to choose from. In their study of digital imaging, Rudnisky et al.
31 chose stereo photography only for views of the disc and macula (ETDRS fields 1 and 2). They used monoscopic photography of the other five (peripheral) fields. This method allowed the investigators some reduction in time, effort, and patient discomfort due to the need for fewer images, while retaining the benefit of the stereo effect for the most critical retinal sites.
Airlie House believed stereo fundus photography would facilitate qualitative and quantitative comparisons of diabetic vascular abnormalities in collaborative studies. They advocated its use whenever possible.
1 The ETDRS stereo protocol was designed to be rigorous, so that even small progressions of diabetic retinopathy could be distinguished in the large number of patients seen in multicenter clinical trials.
27 However, the benefit of grading from stereo versus single photographs was never fully tested. We found that grading a broad range of diabetic retinopathy severity levels by using monoscopic digital images produced results equivalent to using stereo digital images or stereo 35-mm slides. These results suggest that a stereo effect may not be critical for accurate classification of ETDRS diabetic retinopathy severity when using current technology and an optimized framework for fundus photography acquisition and reviewing. If so, the added cost and burden of stereo photography may not be justified.
Supported by a grant from Juvenile Diabetes Foundation Research International, New York, NY (HKL), and by unrestricted grants from Research to Prevent Blindness (Department of Ophthalmology and Visual Sciences, The University of Texas Medical Branch, and the Department of Ophthalmology and Visual Sciences, University of Wisconsin School of Medicine and Public Health).
The authors thank the staff from the Department of Ophthalmology and Visual Sciences, University of Wisconsin-Madison, including the Ocular Epidemiology Group of Barbara E. K. Klein, MD, and Ronald Klein, MD: Andrew F. Ewen, Anne E. Mosher, and Maria K. Swift for grading diabetic retinopathy images; Stacy M. Meuer for grading supervision and Daniel P. Murach for computer support; the Fundus Photograph Reading Center directed by Ronald P. Danis, MD: Trina M. Harding for grading orientation, Qian Peng for statistical advice, Jeff T. Klaves for statistical analyses, and Matthew D. Davis, MD, for suggestions regarding the manuscript.