1040-5488/15/9204-0500/0 VOL. 92, NO. 4, PP. 500Y505 OPTOMETRY AND VISION SCIENCE Copyright * 2015 American Academy of Optometry
ORIGINAL ARTICLE
Comparison of Three Types of Images for the Detection of Retinal Nerve Fiber Layer Defects Hyoung Won Bae*, Naeun Lee*, Chan Yun Kim†, Moonjung Choi*, Samin Hong†, and Gong Je Seong†
ABSTRACT Purpose. To compare the clinical effectiveness of three types of images for detecting retinal nerve fiber layer (RNFL) defects. Methods. Three image sets of 100 subjects (9 normal control subjects, 16 glaucoma suspects, and 75 glaucoma patients) were produced using color fundus photography, typical red-free RNFL photography, and blue reflectance RNFL photography with confocal scanning laser ophthalmoscopy (CSLO). A total of 300 images were rated twice in random order by five independent evaluators who were masked to the patient characteristics; each image was rated as normal, having a diffuse RNFL defect, or showing a wedge RNFL defect. Intraobserver and interobserver agreement, sensitivity, specificity, accuracy, and area under the curve were assessed. An additional analysis was performed for identifying differences in two black-and-white RNFL photographs. Results. The results showed high intraobserver agreement, with relatively low interobserver agreements among the five evaluators. Blue reflectance RNFL photography with CSLO demonstrated the best performance in sensitivity, specificity, accuracy, and area under the curve. Blue reflectance RNFL images showed better accuracy than red-free RNFL images especially in subjects with wedge defects and in advanced glaucomatous cases. Conclusions. The RNFL images produced using blue reflectance with CSLO showed the best performance for the detection of RNFL defects, especially in cases with wedge defects and advanced glaucoma stages. (Optom Vis Sci 2015;92:500Y505) Key Words: retinal nerve fiber layer, red-free photography, blue reflectance, confocal scanning laser ophthalmoscopy
A
n important step in the diagnosis of glaucoma is the evaluation of the retinal nerve fiber layer (RNFL) to identify glaucomatous damage.1Y5 Although rapidly evolving technologies such as optical coherence tomography have been used to evaluate the RNFL, such equipment can be costly and difficult to supply in the clinical setting. Moreover, narrow localized RNFL defects might show false-positive results in some cases. 6,7 For these reasons, typical RNFL photography in conjunction with visual field (VF) tests remains the clinical standard for detecting and evaluating glaucoma in clinical practice. Typically, areas with RNFL defects are less reflective and have dark radiating bands in the arcuate paths of the nerve fiber bundles.3,8Y10 However, in many cases, the border contrast in an
*MD † MD, PhD Department of Ophthalmology, Severance Hospital, Institute of Vision Research, Yonsei University College of Medicine, Seoul, Korea (HWB, CYK, MC, SH, GJS); and Department of Ophthalmology, Hallym Hospital, Incheon, Korea (NL).
RNFL defect appears faint and indistinguishable using standard colored fundus photographic imaging. To enhance the contrast in RNFL defects, red-free illumination can be applied.3 As red-free illumination uses the principle of light reflected from the fundus of the eye, retinas with copious rhodopsin will preferentially absorb blue-green light than red and violet. By selectively illuminating the whitish nerve fiber layer against the red background of the retinal pigment epithelium and choroid, red-free photography can enhance the visibility of the RNFL.11,12 Confocal scanning laser ophthalmoscopy (CSLO) is also another method for assessing the RNFL. The CSLO imaging system uses an optically pumped solid-state laser source to generate the excitation wavelength of 488 nm for blue light reflectance,13,14 and several studies reported that the monochromatic argon blue light of the CSLO is suitable for RNFL evaluation.1,15,16 To the best of our knowledge, no study has compared the utility of each photographic method in evaluating the RNFL in clinical practice. This study compared the effectiveness of three different photographic methods for detecting RNFLs: color fundus photography (CP), typical red-free RNFL photography (RFP), and blue reflectance RNFL photography with CSLO (BRP).
Optometry and Vision Science, Vol. 92, No. 4, April 2015
Copyright © American Academy of Optometry. Unauthorized reproduction of this article is prohibited.
Image Comparisons for Detection of RNFL DefectsVBae et al.
METHODS Subjects We obtained 300 photographs of 100 eyes in 100 consecutive patients who visited the Glaucoma Service Clinic of Yonsei University Medical Center between April 2009 and August 2011 with the following diagnoses: 9 normal control subjects, 16 glaucoma suspects, and 75 glaucoma patients. This study was conducted in accordance with the ethical standards stated in the Declaration of Helsinki and was approved by the institutional review board of the Yonsei University College of Medicine.
Three Types of Images Three image sets of 100 subjects (a total of 300 images) were obtained using CP, RFP, and BRP, and all images were taken after papillary dilation. Color fundus photography and RFP images were taken using a Visucam (Carl Zeiss Meditec, Inc), and RFP images were acquired using green filters with wavelengths of 501 to 580 nm. Blue reflectance RNFL photography with CSLO images were obtained using HRA (Heidelberg Engineering, Heidelberg, Germany) with blue light at a wavelength of 488 nm (Fig. 1).
Evaluation of RNFL Defects Five evaluators (two general ophthalmologists and three ophthalmology trainees) independently rated each of the photographs twice in random order with a 1-week interval between the two examinations. The evaluators were masked to the characteristics, diagnoses, or other clinical information of the patients and classified the RNFLs as having normal anatomy, wedge defects, or diffuse defects. The normal RNFL group was defined as having a normal nerve fiber striation pattern with no dark areas in all peripapillary sectors. The wedge defect group had darker focal areas that originated from the disc border toward the periphery surrounded by normal RNFL striation; its width was larger than that of the retinal vessel at its narrowest. The diffuse defect group was defined as having a relatively faint RNFL defect with an indistinct border. A combination of diffuse and wedge defects was
501
classified as the wedge defect group. The superior and inferior regions were measured independently. Normal and defective RNFLs were defined and standardized via agreement among three glaucoma specialists using all three photographic imaging methods in conjunction with VF and optical coherence tomography results. First, two glaucoma specialists (HWB and NL) examined the 300 images; then, in cases with disagreement, a third glaucoma specialist (CYK) confirmed the discrepancy and discussed the case. After discussion, all discrepancies were settled with a consensus among the three specialists. Agreement was established when both the superior and inferior regions were in accord with the standard agreement at the same time.
Statistical Analysis Kappa statistics were used to assess the intraobserver and interobserver agreement, and the Allan method was used to compare kappa values.17 The sensitivity, specificity, and accuracy were calculated by comparing their values to the standard agreement for evaluating the diagnostic performance of each image. Area under the curve (AUC) was also calculated and compared. To compare the utility of each photographic technique, the McNemar test and the DeLong method were used to assess diagnostic performance and AUC, respectively. In addition, the subjects were divided into two groups based on the difference in accuracy between RFP and BRP: the RFPand the BRP-superior group. The characteristics of the two groups were compared using independent t tests and W2 tests (or Fisher exact test). All statistical analyses were performed using SAS 9.2 (SAS Institute Inc, Cary, NC), and a p value less than 0.05 was considered to indicate significance.
RESULTS Characteristics of Subjects The subjects consisted of 54 men and 46 women; their mean (TSD) age was 55.07 (T14.15) years (range, 20 to 86 years). The mean (TSD) spherical equivalent (SE) of the eyes was j1.80 (T2.68) diopters, and
FIGURE 1. Examples of three types of images assessed by five evaluators: (A) CP, (B) RFP, and (C) BRP. A total of 300 images from 100 subjects were provided to each evaluator in random order. A color version of this figure is available online at www.optvissci.com. Optometry and Vision Science, Vol. 92, No. 4, April 2015
Copyright © American Academy of Optometry. Unauthorized reproduction of this article is prohibited.
502 Image Comparisons for Detection of RNFL DefectsVBae et al. TABLE 1.
Characteristics of the 100 cases assessed by five evaluators Composition of cases* (n = 100) Characteristics SE, D† Axial length, mm† MD, dB† PSD, dB† RNFL defect type, n (%) No visible defect Diffuse defect Wedge defect
Normal (n = 9)
Glaucoma suspect (n = 16)
Mild glaucoma (n = 49)
Moderate glaucoma (n = 13)
Advanced glaucoma (n = 13)
j2.24 T 3.24 24.12 T 1.71 0.11 T 0.85 1.38 T 0.28
j1.49 T 2.81 24.02 T 1.42 j2.51 T 2.06 2.61 T 2.49
j2.10 T 2.74 24.63 T 1.53 j2.68 T 1.66 2.79 T 1.91
j2.05 T 2.83 23.96 T 1.47 j9.29 T 1.71 8.03 T 3.96
j0.50 T 1.38 23.41 T 1.17 j19.81 T 4.96 12.51 T 1.31
9 (100) 0 (0) 0 (0)
9 (56.3) 7 (43.7) 0 (0)
14 (28.6) 17 (34.7) 18 (36.7)
1 (7.7) 5 (38.5) 7 (53.8)
0 3 (23.1) 10 (76.9)
*The stage of glaucoma was graded based on the MD value of the VF. †Data presented as means T SD.
the mean (TSD) axial length was 24.24 (T1.51) mm. The mean deviation (MD) and pattern standard deviation (PSD) of the VF were j5.49 (T6.51) dB and 4.58 (T4.22) dB, respectively. There were 9 normal subjects, 16 glaucoma suspects, and 75 subjects with glaucoma. Glaucoma was categorized into three stages based on the MD value of VF: MD better than j6 dB was considered as mild; MD between j6 and j12 dB was considered as moderate; and MD worse than j12 dB was classified as advanced. Among the 75 glaucoma patients, 49 patients were in a mild stage of disease, 13 were in a moderate stage, and 13 were in an advanced stage. Detailed characteristics of these cases are presented in Table 1.
Diagnostic Performance and AUCs Table 2 shows the diagnostic performance and AUCs among the three photographic methods. The sensitivities/specificities of the CP, RFP, and BRP for the detection of RNFL defects were 67.4%/90.8%, 73.5%/89.4%, and 74.1%/92.2%, respectively. Both RFP and BRP showed significant superiority to CP in sensitivity and accuracy. The specificity was significantly higher in BRP than in RFP. As for the AUCs, BRP showed the highest AUC, followed by RFP and CP (0.83, 0.81, and 0.79, respectively), although a statistical significance was only shown between CP and the other photographs.
Intraobserver and Interobserver Agreements The intraobserver agreements between the first and second tests were almost perfect among all three photographic methods. The kappa coefficient was 0.87 (95% confidence interval [CI], 0.84 to 0.90), 0.86 (95% CI, 0.83 to 0.89), and 0.86 (95% CI, 0.83 to 0.89) for CP, RFP, and BRP, respectively (data not shown). The interobserver agreement among the five evaluators for CP and RFP was almost the same (0.44; 95% CI, 0.40 to 0.48, and 0.41 to 0.48, respectively). However, there was slight improvement (0.48; 95% CI, 0.45 to 0.52) with a nonsignificant trend when using BRP (p = 0.09) (data not shown).
Analysis of Black-and-White Images Obtained via RFP and BRP We additionally analyzed the difference in accuracy between the black-and-white images obtained using RFP and BRP in each subject. Of the total 100 subjects, BRP showed superior accuracy in 42 subjects (BRP-superior group), whereas RFP showed greater accuracy in 27 subjects (RFP-superior group). The two methods had identical accuracy in the remaining 31 subjects. No significant differences were observed between the BRPsuperior and RFP-superior groups in SE and axial length.
TABLE 2.
Diagnostic performance and AUCs for three types of images p Sensitivity, % (95% CI) Specificity, % (95% CI) Accuracy, % (95% CI) AUC (95% CI)
CP (1)
RFP (2)
BRP (3)
(1) vs. (2)
(1) vs. (3)
(2) vs. (3)
67.4 (64.6Y70.1) 90.8 (88.9Y92.7) 77.5 (75.6Y79.3) 0.79 (0.77Y0.81)
73.5 (70.9Y76.1) 89.4 (87.4Y91.5) 80.4 (78.6Y82.1) 0.81 (0.80Y0.83)
74.1 (71.6Y76.7) 92.2 (90.4Y94.0) 81.9 (80.2Y83.6) 0.83 (0.82Y0.85)
G0.0001*
G0.0001*
0.6407*
0.1851*
0.2254*
0.0186*
0.0003*
G0.0001*
0.0874*
0.0023†
G0.0001†
0.0543†
*p values were derived using the McNemar test. †p values were derived using the DeLong method. Optometry and Vision Science, Vol. 92, No. 4, April 2015
Copyright © American Academy of Optometry. Unauthorized reproduction of this article is prohibited.
Image Comparisons for Detection of RNFL DefectsVBae et al.
503
TABLE 3.
Comparison of characteristics of two groups in which each subject showed a difference in accuracy according to the type of image Characteristics SE, D* Axial length, mm* MD, dB* PSD, dB* RNFL defect type, n (%) No visible defect Diffuse defect Wedge defect Glaucoma severity, n (%) Normal Glaucoma suspect Mild glaucoma Moderate glaucoma Advanced glaucoma
RFP-superior group (n = 27)
BRP-superior group (n = 42)
p
j2.37 T 3.04 24.35 T 1.63 j3.23 T 3.05 3.26 T 2.89
j1.47 T 2.59 24.18 T 1.34 j6.48 T 7.28 5.44 T 4.60
9 (33.3) 13 (48.2) 5 (18.5)
10 (23.8) 10 (23.8) 22 (52.4)
4 (14.8) 6 (22.2) 13 (48.2) 4 (14.8) 0 (0)
3 (7.1) 4 (9.5) 22 (52.4) 6 (14.3) 7 (16.7)
0.20† 0.62† 0.01† 0.02† 0.01‡ 0.39‡ 0.04‡ G0.01‡ 0.10‡ 0.42‡ 0.17‡ 0.73‡ 90.99‡ 0.04‡
*Data presented as means T SD. †p values were derived using an independent t test. ‡p values were derived using the W2 test or Fisher exact test.
However, the BRP-superior group had significantly worse MD and PSD values than the RFP-superior group. In addition, the BRPsuperior group showed better performance in detecting wedge RNFL defects than the RFP-superior group, especially in advanced glaucoma cases (Table 3).
DISCUSSION This study compared the clinical performance among three photographic methods (CP, RFP, and BRP) for detecting RNFL
defects. Red-free RNFL photography and BRP, which produce black-and-white images, showed significantly better performance than CP in sensitivity/specificity, accuracy, and AUC. Because early researchers reported that RFP provides higher visibility of the RNFL, it has been well known that RFP detects RNFL defects better than CP.3,12,18 In this study, BRP, which is another black-and-white image, also showed superiority to CP for detecting RNFL defects. However, when RFP and BRP were compared, although there was no statistical significance, BRP showed a tendency of being superior to RFP in all aspects. In addition, BRP showed significantly better
FIGURE 2. Representative case showing the superiority of BRP compared with RFP for the detection of RNFL defects in a case with advanced glaucoma. Blue reflectance RNFL image with CSLO (B) shows more obvious border contrast of the RNFL defect than red-free RNFL image (A) in an advanced glaucoma case in which the MD was j16.75 dB and the PSD was 12.96 dB. Optometry and Vision Science, Vol. 92, No. 4, April 2015
Copyright © American Academy of Optometry. Unauthorized reproduction of this article is prohibited.
504 Image Comparisons for Detection of RNFL DefectsVBae et al.
specificity than RFP. Therefore, our results indicate that RNFL images taken with blue reflectance of CSLO might be the most useful for evaluating RNFL defects. Retinal nerve fiber layer reflectance is used as an RNFL assessment method with CSLO. Retinal nerve fiber layer reflectance is wavelength dependent with higher reflectance at visible wavelengths and lower reflectance at near-infrared wavelengths in the normal retina. Retinal nerve fiber layer reflectance generally arises from light scattering by cylindrical structures in the axons, especially microtubules.19 Thin cylinders contribute more to the reflectance at short wavelengths, whereas thick cylinders dominate the reflectance at long wavelengths. Because glaucomatous damage leads to decreased cylindrical structure in the RNFL, changes in RNFL reflectance at short wavelengths are more sensitive to RNFL damage.20 Therefore, CSLO using the short wavelength of 488 nm facilitates identification of subtle glaucomatous changes in the RNFL and provides a suitable imaging assessment for RNFL defects. Previous studies that assessed the intraobserver and interobserver variability of RNFL evaluation of fundus images detected sizable but tolerable variation in repeatability and reproducibility.10,21 In the present study, however, all images showed high intraobserver agreement but relatively low interobserver agreement. This discrepancy might be because the five evaluators were included in the interobserver agreement assessment, whereas only two tests were used for the intraobserver agreement assessment. Therefore, our results indicate substantial reliability for photographic detection of RNFL defects. To further identify the relative advantage of BRP, we analyzed cases that showed a difference in accuracy between the BRP and RFP images. We found that BRP had the advantage of detecting RNFL defects even when the MD and PSD were worse. Although the RFP images showed limited utility for detecting mild glaucoma with diffuse defects, BRP showed a strong capacity to detect wedge RNFL defects in all stages of glaucoma. Moreover, BRP was far superior to RFP in detecting RNFL defects in advanced glaucoma cases (Fig. 2), likely because BRP has a shorter wavelength than RFP. As mentioned earlier, short wavelengths have higher reflectance in thin microtubules of axons. Therefore, BRP showed higher reflectance than RFP in advanced stages of glaucoma, which demonstrate many thin axons. This study had several limitations. First, it might be possible to infer the RNFL defects from the features of the optic disc when evaluators evaluated each of the photographs. Because a glaucomatous optic disc can show characteristic morphologies such as rim thinning or notching, it could indicate glaucomatous damage if it was not masked in the images. However, the primary purpose of this study was to compare the clinical performance of the different photographic methods, not to evaluate each image directly. Because this issue with the optic disc did not have a large effect on the results of the comparison, our result showing that BRP had the best performance for detecting RNFL defects was unaffected. Second, a selection bias might have existed owing to unclear images or ambiguities in CPs, RFPs, or BRPs. This problem was addressed by including various types of glaucoma subjects in the study to increase generalization. Another limitation was that only images of Korean subjects were examined; whites, for whom the quality of RNFL photography is occasionally
inadequate, were not included. Finally, because of the current limited availability of CSLO, it is clinically difficult to obtain this type of image for the whole fundus area. Although this study showed the superiority of BRP for detecting RNFL defects in clinical practice, it has the disadvantage of reduced availability in many clinics. Future investments should be made to develop new RNFL photographic equipment comparable to CSLO. In conclusion, of the three types of photographic methods, both RFP and BRP showed superior performances to CP in the detection of RNFL defects. However, when those two black-andwhite images were compared, BRP might be superior to RFP for the detection of wedge RNFL defects, especially in cases of advanced glaucoma.
ACKNOWLEDGMENTS The authors are grateful to Hye Sun Lee (Department of Research Affairs, Biostatistics Collaboration Unit, Yonsei University College of Medicine, Seoul, Korea) and Dong-Su Jang (Medical Illustrator, Medical Research Support Section, Yonsei University College of Medicine, Seoul, Korea) for their help with statistical analysis and figures, respectively. This study was supported by a grant from the Korea Health Technology R&D Project, Ministry of Health and Welfare, Republic of Korea (A101727). The authors declare no financial or proprietary interest in the materials or methods mentioned. Received October 7, 2014; accepted December 29, 2014.
REFERENCES 1. Airaksinen PJ, Nieminen H. Retinal nerve fiber layer photography in glaucoma. Ophthalmology 1985;92:877Y9. 2. Sommer A, Katz J, Quigley HA, Miller NR, Robin AL, Richter RC, Witt KA. Clinically detectable nerve fiber atrophy precedes the onset of glaucomatous field loss. Arch Ophthalmol 1991;109:77Y83. 3. Hoyt WF, Frisen L, Newman NM. Fundoscopy of nerve fiber layer defects in glaucoma. Invest Ophthalmol 1973;12:814Y29. 4. Sommer A, Miller NR, Pollack I, Maumenee AE, George T. The nerve fiber layer in the diagnosis of glaucoma. Arch Ophthalmol 1977;95:2149Y56. 5. Quigley HA, Miller NR, George T. Clinical evaluation of nerve fiber layer atrophy as an indicator of glaucomatous optic nerve damage. Arch Ophthalmol 1980;98:1564Y71. 6. Kim NR, Lee ES, Seong GJ, Choi EH, Hong S, Kim CY. Spectraldomain optical coherence tomography for detection of localized retinal nerve fiber layer defects in patients with open-angle glaucoma. Arch Ophthalmol 2010;128:1121Y8. 7. Kim NR, Lim H, Kim JH, Rho SS, Seong GJ, Kim CY. Factors associated with false positives in retinal nerve fiber layer color codes from spectral-domain optical coherence tomography. Ophthalmology 2011;118:1774Y81. 8. Townsend KA, Wollstein G, Schuman JS. Imaging of the retinal nerve fibre layer for glaucoma. Br J Ophthalmol 2009;93:139Y43. 9. Jonas JB, Schiro D. Localised wedge shaped defects of the retinal nerve fibre layer in glaucoma. Br J Ophthalmol 1994;78:285Y90. 10. Airaksinen PJ, Drance SM, Douglas GR, Mawson DK, Nieminen H. Diffuse and localized nerve fiber loss in glaucoma. Am J Ophthalmol 1984;98:566Y71. 11. Mizuno K, Majima A, Ozawa K, Ito H. Red-free light fundus photography. Photographic optogram. Invest Ophthalmol 1968; 7:241Y9.
Optometry and Vision Science, Vol. 92, No. 4, April 2015
Copyright © American Academy of Optometry. Unauthorized reproduction of this article is prohibited.
Image Comparisons for Detection of RNFL DefectsVBae et al. 12. Sommer A, D’Anna SA, Kues HA, George T. High-resolution photography of the retinal nerve fiber layer. Am J Ophthalmol 1983;96:535Y9. 13. Helb HM, Charbel Issa P, Fleckenstein M, Schmitz-Valckenberg S, Scholl HP, Meyer CH, Eter N, Holz FG. Clinical evaluation of simultaneous confocal scanning laser ophthalmoscopy imaging combined with high-resolution, spectral-domain optical coherence tomography. Acta Ophthalmol 2010;88:842Y9. 14. Hassenstein A, Meyer CH. Clinical use and research applications of Heidelberg retinal angiography and spectral-domain optical coherence tomographyVa review. Clin Experiment Ophthalmol 2009; 37:130Y43. 15. Miglior S, Rossetti L, Brigatti L, Bujtar E, Orzalesi N. Reproducibility of retinal nerve fiber layer evaluation by dynamics scanning laser ophthalmoscopy. Am J Ophthalmol 1994;118:16Y23. 16. Behrendt T, Wilson LA. Spectral reflectance photography of the retina. Am J Ophthalmol 1965;59:1079Y88. 17. Donner A, Shoukri MM, Klar N, Bartfay E. Testing the equality of two dependent kappa statistics. Stat Med 2000;19:373Y87.
505
18. Fulk GW, Van Veen HG. How to photograph and evaluate the retinal nerve fiber layer. J Am Optom Assoc 1986;57:760Y3. 19. Knighton RW, Huang XR. Directional and spectral reflectance of the rat retinal nerve fiber layer. Invest Ophthalmol Vis Sci 1999; 40:639Y47. 20. Huang XR, Zhou Y, Knighton RW, Kong W, Feuer WJ. Wavelength-dependent change of retinal nerve fiber layer reflectance in glaucomatous retinas. Invest Ophthalmol Vis Sci 2012;53:5869Y76. 21. Sommer A, Quigley HA, Robin AL, Miller NR, Katz J, Arkell S. Evaluation of nerve fiber layer assessment. Arch Ophthalmol 1984;102:1766Y71.
Gong Je Seong Department of Ophthalmology Yonsei University College of Medicine 250 Seongsanno, Seodaemun-gu Seoul 120-752 Republic of Korea e-mail: [email protected]
Optometry and Vision Science, Vol. 92, No. 4, April 2015
Copyright © American Academy of Optometry. Unauthorized reproduction of this article is prohibited.
ORIGINAL ARTICLE
Comparison of Three Types of Images for the Detection of Retinal Nerve Fiber Layer Defects Hyoung Won Bae*, Naeun Lee*, Chan Yun Kim†, Moonjung Choi*, Samin Hong†, and Gong Je Seong†
ABSTRACT Purpose. To compare the clinical effectiveness of three types of images for detecting retinal nerve fiber layer (RNFL) defects. Methods. Three image sets of 100 subjects (9 normal control subjects, 16 glaucoma suspects, and 75 glaucoma patients) were produced using color fundus photography, typical red-free RNFL photography, and blue reflectance RNFL photography with confocal scanning laser ophthalmoscopy (CSLO). A total of 300 images were rated twice in random order by five independent evaluators who were masked to the patient characteristics; each image was rated as normal, having a diffuse RNFL defect, or showing a wedge RNFL defect. Intraobserver and interobserver agreement, sensitivity, specificity, accuracy, and area under the curve were assessed. An additional analysis was performed for identifying differences in two black-and-white RNFL photographs. Results. The results showed high intraobserver agreement, with relatively low interobserver agreements among the five evaluators. Blue reflectance RNFL photography with CSLO demonstrated the best performance in sensitivity, specificity, accuracy, and area under the curve. Blue reflectance RNFL images showed better accuracy than red-free RNFL images especially in subjects with wedge defects and in advanced glaucomatous cases. Conclusions. The RNFL images produced using blue reflectance with CSLO showed the best performance for the detection of RNFL defects, especially in cases with wedge defects and advanced glaucoma stages. (Optom Vis Sci 2015;92:500Y505) Key Words: retinal nerve fiber layer, red-free photography, blue reflectance, confocal scanning laser ophthalmoscopy
A
n important step in the diagnosis of glaucoma is the evaluation of the retinal nerve fiber layer (RNFL) to identify glaucomatous damage.1Y5 Although rapidly evolving technologies such as optical coherence tomography have been used to evaluate the RNFL, such equipment can be costly and difficult to supply in the clinical setting. Moreover, narrow localized RNFL defects might show false-positive results in some cases. 6,7 For these reasons, typical RNFL photography in conjunction with visual field (VF) tests remains the clinical standard for detecting and evaluating glaucoma in clinical practice. Typically, areas with RNFL defects are less reflective and have dark radiating bands in the arcuate paths of the nerve fiber bundles.3,8Y10 However, in many cases, the border contrast in an
*MD † MD, PhD Department of Ophthalmology, Severance Hospital, Institute of Vision Research, Yonsei University College of Medicine, Seoul, Korea (HWB, CYK, MC, SH, GJS); and Department of Ophthalmology, Hallym Hospital, Incheon, Korea (NL).
RNFL defect appears faint and indistinguishable using standard colored fundus photographic imaging. To enhance the contrast in RNFL defects, red-free illumination can be applied.3 As red-free illumination uses the principle of light reflected from the fundus of the eye, retinas with copious rhodopsin will preferentially absorb blue-green light than red and violet. By selectively illuminating the whitish nerve fiber layer against the red background of the retinal pigment epithelium and choroid, red-free photography can enhance the visibility of the RNFL.11,12 Confocal scanning laser ophthalmoscopy (CSLO) is also another method for assessing the RNFL. The CSLO imaging system uses an optically pumped solid-state laser source to generate the excitation wavelength of 488 nm for blue light reflectance,13,14 and several studies reported that the monochromatic argon blue light of the CSLO is suitable for RNFL evaluation.1,15,16 To the best of our knowledge, no study has compared the utility of each photographic method in evaluating the RNFL in clinical practice. This study compared the effectiveness of three different photographic methods for detecting RNFLs: color fundus photography (CP), typical red-free RNFL photography (RFP), and blue reflectance RNFL photography with CSLO (BRP).
Optometry and Vision Science, Vol. 92, No. 4, April 2015
Copyright © American Academy of Optometry. Unauthorized reproduction of this article is prohibited.
Image Comparisons for Detection of RNFL DefectsVBae et al.
METHODS Subjects We obtained 300 photographs of 100 eyes in 100 consecutive patients who visited the Glaucoma Service Clinic of Yonsei University Medical Center between April 2009 and August 2011 with the following diagnoses: 9 normal control subjects, 16 glaucoma suspects, and 75 glaucoma patients. This study was conducted in accordance with the ethical standards stated in the Declaration of Helsinki and was approved by the institutional review board of the Yonsei University College of Medicine.
Three Types of Images Three image sets of 100 subjects (a total of 300 images) were obtained using CP, RFP, and BRP, and all images were taken after papillary dilation. Color fundus photography and RFP images were taken using a Visucam (Carl Zeiss Meditec, Inc), and RFP images were acquired using green filters with wavelengths of 501 to 580 nm. Blue reflectance RNFL photography with CSLO images were obtained using HRA (Heidelberg Engineering, Heidelberg, Germany) with blue light at a wavelength of 488 nm (Fig. 1).
Evaluation of RNFL Defects Five evaluators (two general ophthalmologists and three ophthalmology trainees) independently rated each of the photographs twice in random order with a 1-week interval between the two examinations. The evaluators were masked to the characteristics, diagnoses, or other clinical information of the patients and classified the RNFLs as having normal anatomy, wedge defects, or diffuse defects. The normal RNFL group was defined as having a normal nerve fiber striation pattern with no dark areas in all peripapillary sectors. The wedge defect group had darker focal areas that originated from the disc border toward the periphery surrounded by normal RNFL striation; its width was larger than that of the retinal vessel at its narrowest. The diffuse defect group was defined as having a relatively faint RNFL defect with an indistinct border. A combination of diffuse and wedge defects was
501
classified as the wedge defect group. The superior and inferior regions were measured independently. Normal and defective RNFLs were defined and standardized via agreement among three glaucoma specialists using all three photographic imaging methods in conjunction with VF and optical coherence tomography results. First, two glaucoma specialists (HWB and NL) examined the 300 images; then, in cases with disagreement, a third glaucoma specialist (CYK) confirmed the discrepancy and discussed the case. After discussion, all discrepancies were settled with a consensus among the three specialists. Agreement was established when both the superior and inferior regions were in accord with the standard agreement at the same time.
Statistical Analysis Kappa statistics were used to assess the intraobserver and interobserver agreement, and the Allan method was used to compare kappa values.17 The sensitivity, specificity, and accuracy were calculated by comparing their values to the standard agreement for evaluating the diagnostic performance of each image. Area under the curve (AUC) was also calculated and compared. To compare the utility of each photographic technique, the McNemar test and the DeLong method were used to assess diagnostic performance and AUC, respectively. In addition, the subjects were divided into two groups based on the difference in accuracy between RFP and BRP: the RFPand the BRP-superior group. The characteristics of the two groups were compared using independent t tests and W2 tests (or Fisher exact test). All statistical analyses were performed using SAS 9.2 (SAS Institute Inc, Cary, NC), and a p value less than 0.05 was considered to indicate significance.
RESULTS Characteristics of Subjects The subjects consisted of 54 men and 46 women; their mean (TSD) age was 55.07 (T14.15) years (range, 20 to 86 years). The mean (TSD) spherical equivalent (SE) of the eyes was j1.80 (T2.68) diopters, and
FIGURE 1. Examples of three types of images assessed by five evaluators: (A) CP, (B) RFP, and (C) BRP. A total of 300 images from 100 subjects were provided to each evaluator in random order. A color version of this figure is available online at www.optvissci.com. Optometry and Vision Science, Vol. 92, No. 4, April 2015
Copyright © American Academy of Optometry. Unauthorized reproduction of this article is prohibited.
502 Image Comparisons for Detection of RNFL DefectsVBae et al. TABLE 1.
Characteristics of the 100 cases assessed by five evaluators Composition of cases* (n = 100) Characteristics SE, D† Axial length, mm† MD, dB† PSD, dB† RNFL defect type, n (%) No visible defect Diffuse defect Wedge defect
Normal (n = 9)
Glaucoma suspect (n = 16)
Mild glaucoma (n = 49)
Moderate glaucoma (n = 13)
Advanced glaucoma (n = 13)
j2.24 T 3.24 24.12 T 1.71 0.11 T 0.85 1.38 T 0.28
j1.49 T 2.81 24.02 T 1.42 j2.51 T 2.06 2.61 T 2.49
j2.10 T 2.74 24.63 T 1.53 j2.68 T 1.66 2.79 T 1.91
j2.05 T 2.83 23.96 T 1.47 j9.29 T 1.71 8.03 T 3.96
j0.50 T 1.38 23.41 T 1.17 j19.81 T 4.96 12.51 T 1.31
9 (100) 0 (0) 0 (0)
9 (56.3) 7 (43.7) 0 (0)
14 (28.6) 17 (34.7) 18 (36.7)
1 (7.7) 5 (38.5) 7 (53.8)
0 3 (23.1) 10 (76.9)
*The stage of glaucoma was graded based on the MD value of the VF. †Data presented as means T SD.
the mean (TSD) axial length was 24.24 (T1.51) mm. The mean deviation (MD) and pattern standard deviation (PSD) of the VF were j5.49 (T6.51) dB and 4.58 (T4.22) dB, respectively. There were 9 normal subjects, 16 glaucoma suspects, and 75 subjects with glaucoma. Glaucoma was categorized into three stages based on the MD value of VF: MD better than j6 dB was considered as mild; MD between j6 and j12 dB was considered as moderate; and MD worse than j12 dB was classified as advanced. Among the 75 glaucoma patients, 49 patients were in a mild stage of disease, 13 were in a moderate stage, and 13 were in an advanced stage. Detailed characteristics of these cases are presented in Table 1.
Diagnostic Performance and AUCs Table 2 shows the diagnostic performance and AUCs among the three photographic methods. The sensitivities/specificities of the CP, RFP, and BRP for the detection of RNFL defects were 67.4%/90.8%, 73.5%/89.4%, and 74.1%/92.2%, respectively. Both RFP and BRP showed significant superiority to CP in sensitivity and accuracy. The specificity was significantly higher in BRP than in RFP. As for the AUCs, BRP showed the highest AUC, followed by RFP and CP (0.83, 0.81, and 0.79, respectively), although a statistical significance was only shown between CP and the other photographs.
Intraobserver and Interobserver Agreements The intraobserver agreements between the first and second tests were almost perfect among all three photographic methods. The kappa coefficient was 0.87 (95% confidence interval [CI], 0.84 to 0.90), 0.86 (95% CI, 0.83 to 0.89), and 0.86 (95% CI, 0.83 to 0.89) for CP, RFP, and BRP, respectively (data not shown). The interobserver agreement among the five evaluators for CP and RFP was almost the same (0.44; 95% CI, 0.40 to 0.48, and 0.41 to 0.48, respectively). However, there was slight improvement (0.48; 95% CI, 0.45 to 0.52) with a nonsignificant trend when using BRP (p = 0.09) (data not shown).
Analysis of Black-and-White Images Obtained via RFP and BRP We additionally analyzed the difference in accuracy between the black-and-white images obtained using RFP and BRP in each subject. Of the total 100 subjects, BRP showed superior accuracy in 42 subjects (BRP-superior group), whereas RFP showed greater accuracy in 27 subjects (RFP-superior group). The two methods had identical accuracy in the remaining 31 subjects. No significant differences were observed between the BRPsuperior and RFP-superior groups in SE and axial length.
TABLE 2.
Diagnostic performance and AUCs for three types of images p Sensitivity, % (95% CI) Specificity, % (95% CI) Accuracy, % (95% CI) AUC (95% CI)
CP (1)
RFP (2)
BRP (3)
(1) vs. (2)
(1) vs. (3)
(2) vs. (3)
67.4 (64.6Y70.1) 90.8 (88.9Y92.7) 77.5 (75.6Y79.3) 0.79 (0.77Y0.81)
73.5 (70.9Y76.1) 89.4 (87.4Y91.5) 80.4 (78.6Y82.1) 0.81 (0.80Y0.83)
74.1 (71.6Y76.7) 92.2 (90.4Y94.0) 81.9 (80.2Y83.6) 0.83 (0.82Y0.85)
G0.0001*
G0.0001*
0.6407*
0.1851*
0.2254*
0.0186*
0.0003*
G0.0001*
0.0874*
0.0023†
G0.0001†
0.0543†
*p values were derived using the McNemar test. †p values were derived using the DeLong method. Optometry and Vision Science, Vol. 92, No. 4, April 2015
Copyright © American Academy of Optometry. Unauthorized reproduction of this article is prohibited.
Image Comparisons for Detection of RNFL DefectsVBae et al.
503
TABLE 3.
Comparison of characteristics of two groups in which each subject showed a difference in accuracy according to the type of image Characteristics SE, D* Axial length, mm* MD, dB* PSD, dB* RNFL defect type, n (%) No visible defect Diffuse defect Wedge defect Glaucoma severity, n (%) Normal Glaucoma suspect Mild glaucoma Moderate glaucoma Advanced glaucoma
RFP-superior group (n = 27)
BRP-superior group (n = 42)
p
j2.37 T 3.04 24.35 T 1.63 j3.23 T 3.05 3.26 T 2.89
j1.47 T 2.59 24.18 T 1.34 j6.48 T 7.28 5.44 T 4.60
9 (33.3) 13 (48.2) 5 (18.5)
10 (23.8) 10 (23.8) 22 (52.4)
4 (14.8) 6 (22.2) 13 (48.2) 4 (14.8) 0 (0)
3 (7.1) 4 (9.5) 22 (52.4) 6 (14.3) 7 (16.7)
0.20† 0.62† 0.01† 0.02† 0.01‡ 0.39‡ 0.04‡ G0.01‡ 0.10‡ 0.42‡ 0.17‡ 0.73‡ 90.99‡ 0.04‡
*Data presented as means T SD. †p values were derived using an independent t test. ‡p values were derived using the W2 test or Fisher exact test.
However, the BRP-superior group had significantly worse MD and PSD values than the RFP-superior group. In addition, the BRPsuperior group showed better performance in detecting wedge RNFL defects than the RFP-superior group, especially in advanced glaucoma cases (Table 3).
DISCUSSION This study compared the clinical performance among three photographic methods (CP, RFP, and BRP) for detecting RNFL
defects. Red-free RNFL photography and BRP, which produce black-and-white images, showed significantly better performance than CP in sensitivity/specificity, accuracy, and AUC. Because early researchers reported that RFP provides higher visibility of the RNFL, it has been well known that RFP detects RNFL defects better than CP.3,12,18 In this study, BRP, which is another black-and-white image, also showed superiority to CP for detecting RNFL defects. However, when RFP and BRP were compared, although there was no statistical significance, BRP showed a tendency of being superior to RFP in all aspects. In addition, BRP showed significantly better
FIGURE 2. Representative case showing the superiority of BRP compared with RFP for the detection of RNFL defects in a case with advanced glaucoma. Blue reflectance RNFL image with CSLO (B) shows more obvious border contrast of the RNFL defect than red-free RNFL image (A) in an advanced glaucoma case in which the MD was j16.75 dB and the PSD was 12.96 dB. Optometry and Vision Science, Vol. 92, No. 4, April 2015
Copyright © American Academy of Optometry. Unauthorized reproduction of this article is prohibited.
504 Image Comparisons for Detection of RNFL DefectsVBae et al.
specificity than RFP. Therefore, our results indicate that RNFL images taken with blue reflectance of CSLO might be the most useful for evaluating RNFL defects. Retinal nerve fiber layer reflectance is used as an RNFL assessment method with CSLO. Retinal nerve fiber layer reflectance is wavelength dependent with higher reflectance at visible wavelengths and lower reflectance at near-infrared wavelengths in the normal retina. Retinal nerve fiber layer reflectance generally arises from light scattering by cylindrical structures in the axons, especially microtubules.19 Thin cylinders contribute more to the reflectance at short wavelengths, whereas thick cylinders dominate the reflectance at long wavelengths. Because glaucomatous damage leads to decreased cylindrical structure in the RNFL, changes in RNFL reflectance at short wavelengths are more sensitive to RNFL damage.20 Therefore, CSLO using the short wavelength of 488 nm facilitates identification of subtle glaucomatous changes in the RNFL and provides a suitable imaging assessment for RNFL defects. Previous studies that assessed the intraobserver and interobserver variability of RNFL evaluation of fundus images detected sizable but tolerable variation in repeatability and reproducibility.10,21 In the present study, however, all images showed high intraobserver agreement but relatively low interobserver agreement. This discrepancy might be because the five evaluators were included in the interobserver agreement assessment, whereas only two tests were used for the intraobserver agreement assessment. Therefore, our results indicate substantial reliability for photographic detection of RNFL defects. To further identify the relative advantage of BRP, we analyzed cases that showed a difference in accuracy between the BRP and RFP images. We found that BRP had the advantage of detecting RNFL defects even when the MD and PSD were worse. Although the RFP images showed limited utility for detecting mild glaucoma with diffuse defects, BRP showed a strong capacity to detect wedge RNFL defects in all stages of glaucoma. Moreover, BRP was far superior to RFP in detecting RNFL defects in advanced glaucoma cases (Fig. 2), likely because BRP has a shorter wavelength than RFP. As mentioned earlier, short wavelengths have higher reflectance in thin microtubules of axons. Therefore, BRP showed higher reflectance than RFP in advanced stages of glaucoma, which demonstrate many thin axons. This study had several limitations. First, it might be possible to infer the RNFL defects from the features of the optic disc when evaluators evaluated each of the photographs. Because a glaucomatous optic disc can show characteristic morphologies such as rim thinning or notching, it could indicate glaucomatous damage if it was not masked in the images. However, the primary purpose of this study was to compare the clinical performance of the different photographic methods, not to evaluate each image directly. Because this issue with the optic disc did not have a large effect on the results of the comparison, our result showing that BRP had the best performance for detecting RNFL defects was unaffected. Second, a selection bias might have existed owing to unclear images or ambiguities in CPs, RFPs, or BRPs. This problem was addressed by including various types of glaucoma subjects in the study to increase generalization. Another limitation was that only images of Korean subjects were examined; whites, for whom the quality of RNFL photography is occasionally
inadequate, were not included. Finally, because of the current limited availability of CSLO, it is clinically difficult to obtain this type of image for the whole fundus area. Although this study showed the superiority of BRP for detecting RNFL defects in clinical practice, it has the disadvantage of reduced availability in many clinics. Future investments should be made to develop new RNFL photographic equipment comparable to CSLO. In conclusion, of the three types of photographic methods, both RFP and BRP showed superior performances to CP in the detection of RNFL defects. However, when those two black-andwhite images were compared, BRP might be superior to RFP for the detection of wedge RNFL defects, especially in cases of advanced glaucoma.
ACKNOWLEDGMENTS The authors are grateful to Hye Sun Lee (Department of Research Affairs, Biostatistics Collaboration Unit, Yonsei University College of Medicine, Seoul, Korea) and Dong-Su Jang (Medical Illustrator, Medical Research Support Section, Yonsei University College of Medicine, Seoul, Korea) for their help with statistical analysis and figures, respectively. This study was supported by a grant from the Korea Health Technology R&D Project, Ministry of Health and Welfare, Republic of Korea (A101727). The authors declare no financial or proprietary interest in the materials or methods mentioned. Received October 7, 2014; accepted December 29, 2014.
REFERENCES 1. Airaksinen PJ, Nieminen H. Retinal nerve fiber layer photography in glaucoma. Ophthalmology 1985;92:877Y9. 2. Sommer A, Katz J, Quigley HA, Miller NR, Robin AL, Richter RC, Witt KA. Clinically detectable nerve fiber atrophy precedes the onset of glaucomatous field loss. Arch Ophthalmol 1991;109:77Y83. 3. Hoyt WF, Frisen L, Newman NM. Fundoscopy of nerve fiber layer defects in glaucoma. Invest Ophthalmol 1973;12:814Y29. 4. Sommer A, Miller NR, Pollack I, Maumenee AE, George T. The nerve fiber layer in the diagnosis of glaucoma. Arch Ophthalmol 1977;95:2149Y56. 5. Quigley HA, Miller NR, George T. Clinical evaluation of nerve fiber layer atrophy as an indicator of glaucomatous optic nerve damage. Arch Ophthalmol 1980;98:1564Y71. 6. Kim NR, Lee ES, Seong GJ, Choi EH, Hong S, Kim CY. Spectraldomain optical coherence tomography for detection of localized retinal nerve fiber layer defects in patients with open-angle glaucoma. Arch Ophthalmol 2010;128:1121Y8. 7. Kim NR, Lim H, Kim JH, Rho SS, Seong GJ, Kim CY. Factors associated with false positives in retinal nerve fiber layer color codes from spectral-domain optical coherence tomography. Ophthalmology 2011;118:1774Y81. 8. Townsend KA, Wollstein G, Schuman JS. Imaging of the retinal nerve fibre layer for glaucoma. Br J Ophthalmol 2009;93:139Y43. 9. Jonas JB, Schiro D. Localised wedge shaped defects of the retinal nerve fibre layer in glaucoma. Br J Ophthalmol 1994;78:285Y90. 10. Airaksinen PJ, Drance SM, Douglas GR, Mawson DK, Nieminen H. Diffuse and localized nerve fiber loss in glaucoma. Am J Ophthalmol 1984;98:566Y71. 11. Mizuno K, Majima A, Ozawa K, Ito H. Red-free light fundus photography. Photographic optogram. Invest Ophthalmol 1968; 7:241Y9.
Optometry and Vision Science, Vol. 92, No. 4, April 2015
Copyright © American Academy of Optometry. Unauthorized reproduction of this article is prohibited.
Image Comparisons for Detection of RNFL DefectsVBae et al. 12. Sommer A, D’Anna SA, Kues HA, George T. High-resolution photography of the retinal nerve fiber layer. Am J Ophthalmol 1983;96:535Y9. 13. Helb HM, Charbel Issa P, Fleckenstein M, Schmitz-Valckenberg S, Scholl HP, Meyer CH, Eter N, Holz FG. Clinical evaluation of simultaneous confocal scanning laser ophthalmoscopy imaging combined with high-resolution, spectral-domain optical coherence tomography. Acta Ophthalmol 2010;88:842Y9. 14. Hassenstein A, Meyer CH. Clinical use and research applications of Heidelberg retinal angiography and spectral-domain optical coherence tomographyVa review. Clin Experiment Ophthalmol 2009; 37:130Y43. 15. Miglior S, Rossetti L, Brigatti L, Bujtar E, Orzalesi N. Reproducibility of retinal nerve fiber layer evaluation by dynamics scanning laser ophthalmoscopy. Am J Ophthalmol 1994;118:16Y23. 16. Behrendt T, Wilson LA. Spectral reflectance photography of the retina. Am J Ophthalmol 1965;59:1079Y88. 17. Donner A, Shoukri MM, Klar N, Bartfay E. Testing the equality of two dependent kappa statistics. Stat Med 2000;19:373Y87.
505
18. Fulk GW, Van Veen HG. How to photograph and evaluate the retinal nerve fiber layer. J Am Optom Assoc 1986;57:760Y3. 19. Knighton RW, Huang XR. Directional and spectral reflectance of the rat retinal nerve fiber layer. Invest Ophthalmol Vis Sci 1999; 40:639Y47. 20. Huang XR, Zhou Y, Knighton RW, Kong W, Feuer WJ. Wavelength-dependent change of retinal nerve fiber layer reflectance in glaucomatous retinas. Invest Ophthalmol Vis Sci 2012;53:5869Y76. 21. Sommer A, Quigley HA, Robin AL, Miller NR, Katz J, Arkell S. Evaluation of nerve fiber layer assessment. Arch Ophthalmol 1984;102:1766Y71.
Gong Je Seong Department of Ophthalmology Yonsei University College of Medicine 250 Seongsanno, Seodaemun-gu Seoul 120-752 Republic of Korea e-mail: [email protected]
Optometry and Vision Science, Vol. 92, No. 4, April 2015
Copyright © American Academy of Optometry. Unauthorized reproduction of this article is prohibited.
