Graefe's Archive
Graefe's Arch Clin Exp Ophthalmol (1992)230:432-436
for Cgnical and Experimental
Ophthalmology © Springer-Verlag 1992
Digital image analysis of optic nerve head pallor as a diagnostic test for early glaucoma* Albert Assad and Joseph Caprioli Glaucoma Section, Yale University School of Medicine, Department of Ophthalmology and Visual Science, 330 Cedar Street, New Haven, CT 06510, USA Received October 14, 1991 /Accepted December 23, 1991
Abstract. We developed a computer-based technique to quantify optic nerve head pallor from videographically acquired digitized optic nerve images and tested the ability of pallor measurements to discriminate between normal eyes and eyes with early glaucoma. Corresponding pixel values from images obtained under 540 nm (green) and 640 nm (red) light with a videographic fundus camera were used to quantify optic nerve head pallor. A pallor density histogram was calculated for each eye, and contained values betwen 0 (red) to 1 (white). A measure o f the distribution width o f the histogram provided pallor measurements standardized to the measurements o f the large veins o f the disc. A database o f one eye each of 44 normal controls and 70 patients with early open angle glaucoma was used to test the measurements for diagnostic sensitivity and specificity. These standardized pallor measurements did not perform better than absolute pallor measurements to discriminate between normal and glaucomatous eyes. The sensitivity and specificity of standardized pallor measurements (49% and 57%, respectively, for this database) were not as good as those for stereoscopic measurements o f disc rim area in the same database (70% and 73%). Pallor measurements of this type do not appear to be sensitive or specific indicators of early glaucoma.
Introduction The poor ability o f intraocular pressure measurements to differentiate between normal and glaucomatous eyes and the subjectivity and variability of visual field measurements have placed additional emphasis on objective measurements of physical attributes of the optic disc * Supported in part by grants from Research to Prevent Blindness, Inc, The Connecticut Lions Eye Research Foundation, and Alcon Laboratories, Inc. ; the authors have no proprietary interest in the instruments or softwa/e described within
Correspondence to: J. Caprioli
and nerve fiber layer to detect early glaucoma. Decisions about diagnosis or treatment are often based on a change in the appearance o f the optic disc during the course o f the disease. Pallor o f the optic disc is considered an important sign o f optic nerve atrophy and is a manifestation of the loss o f vascularized neural tissue. This study was performed to evaluate the ability of pallor measurements derived from digital videographic imaging to differentiate between normal eyes and those with early glaucoma.
Materials and methods
Subjects A database was constructed from images of 44 normal subjects and 70 glaucoma patients. The glaucoma patients were agematched to the normal subjects. Normal subjects were recruited from hospital staff and friends or spouses of patients; none had presented for medical evaluation. Patients were considered glaucomatous if they had elevated intraocular pressure or a history of elevated intraocular pressure before treatment, and early to moderate, typical glaucomatous visual field defects. Typical glaucomatous visual field defects were defined, in a reliably performed automated threshold visual field test, as at least: 1. Two or more contiguous points with 10 dB loss or greater in the superior or inferior Bjerrum areas, compared with perimeterdefined age-matched controls 2. Three or more contiguous points with 5 dB loss or greater in the superior or inferior Bjerrum areas 3. A 10 dB difference across the nasal horizontal midline in two or more adjacent locations. The most superior and inferior rows of threshold measurements from the 30° programs were excluded from these criteria to avoid the inclusion of rim artifacts. Visual field indices mean defect, corrected loss variance, and short term fluctuation as described by Flammer et al. were calculated for each visual field [5].
Image analysis Image acquisition of the optic disc was performed with a videographic instrument designed for this purpose (Rodenstock Instru-
433 ments, Munich, FRG). We developed software, which ran on a stand-alone PC to construct pallor density histograms from the digitized images. The reproducibility of these measurements has previously been reported [8]. The digitized images stored by the videographic system consist of a matrix of 256 x 256 points. Each pixel has a value eight bits deep; values ranged from 0-255. The corresponding pixel values from green and red monochromatic images obtained with a videographic fundus camera were used to calculate the pallor of the optic nerve head according to equation 1. The green digitized image was obtained under 540 nm light and the red was obtained under 640 nm light. The pallor density of each pixel was defined as twice the green reflectance divided by the sum of the green and red reflectance: 2 x (green reflectance) pallor density - (red reflectance) + (green reflectance)
(1)
The selection of Eq. (1) for calculating pallor was based on the model described by Davies [4]. In this model, a grey-white nerve head, which is equally reflective to both red and green light, is covered by a thin layer of blood. The relative blood layer thickness (RBLT) is proportional to the difference of the log of the reflectances at green and red wavelength: RBLT = log (green reflectance) - log (red reflectance) + 1
To compare different test parameters, receiver-operator characteristic (ROC) curves were used. ROC curves display the entire range of sensitivity and specificity values and provide a means for evaluating whether a certain test or parameter is better than another at detecting differences between groups. The greater the area under the curve, the better the test is at discriminating between the normal and glaucomatous groups.
Results
Summary data for age, sex, refractive error, visual field indices, and rim area of the optic nerve head are given
Table 1. Descriptive statistics for age, sex, refractive error, visual field, and optic nerve head* Normal
Glaucoma
Males Females
14 30
35 35
Age (years)
59.48 -+ 11.56
60.04_+ 10.53
0.73_+ 1.70
-1.09_+ 3.18
-0.62_+ 1.62 2.20_+ 2.87
7.11_+ 5.79 35.89_+36.31
1.07-t- 0.20
0.82_+ 0.30
(2)
A value of 1 is added to provide an appropriate range of numbers : these range from 0 for a thick blood layer to 1 for the presence of no blood. Equation (1) is an approximation good to within 4% of Eq. (2) for pallor values
0.4
CC
hi
h6
O II.
Ill 15% I
°.
I PALLOR
RED
15%
30%
I
0.2
I
k"
VALUE WHITE
Fig. 1. This figure shows the technique of using trimmed means for describing non-Gaussian distributions. HI is the average pallor value for the whitest 5% of the points and h6 is the average pallor value for the reddest 5% of the points. H2-h5 is the difference between the trimmed means h2 and h5
0.0
:
0.0
I
0.2
FALSE POSITIVE
I
I
0.4 RATE
0.6
I
0.8
1.0
(1-specificity)
Fig. 2. This figure shows the ROC curve for the trimmed mean hl ; hl represents an absolute pallor value
434
1.0
1,0
I
|
=
"~ 0 . 8 c o
__l:
. . . . . .
I
__
0.8
Ul 1-.
0.4
1--
,•
U) 0
I
0.6
0.6 MI
,.
i
0.2
0.2
..
•
•
00o
w er
k-
I-
0.0
r0.0
,
,
0.2
j
0.4
F A L S E P O S I T I V E RATE
0.0
,
0.6
0.8
1.0
I
I
I
, 0.2
J 0.4
FALSE P O S I T I V E RATE
(1-specificity)
Fig, 3. This figure shows an ROC curve for the difference between trimmed mean parameters h2 and h5. H2-h5 is a measure of the width of the pallor frequency distribution
60
0.0
j 0.6
I
0.8
1.0
(1 -specificity)
Fig. 4. An ROC curve for the structural parameter rim area is shown for comparison. This parameter more efficienty discriminated between normals and glaucomas than did the pallor measurements
I
50
lid
r3 < :f i \ v)
,.J ILl uX L
40
30
L
•
'~
/
20
#
,
to
It
t
Fig. 5. The average pallor value distributions for the n o r m a l subjects and glaucoma patients. There is a slight shift t o w a r d increased pallor values in the glaucoma patients, but this was n o t a statistically significant difference. - - N o r m a l Subjects . . - G l a u c o m a Patients
L
10
0
I
0.0
red
0.2
i
i
0.4
0.6
PALLOR
I
0.8
1 .(
white
in Table 1. There was no significant difference between the mean ages of the two groups. Figure 2 is an R O C curve for the non-standardized test parameter hi. H 1 represents the average pallor value of the most pale 5% of the image pixels. This parameter was unable to differentiate well between normal subjects and glaucomatous patients as evidenced by the relatively diagonal line in the R O C curve. The difference between trimmed mean parameters h2 and h5 reflects the distri-
bution width, and is independent of the position of the distribution. This standardization did not improve the ability to discriminate between these normal and glaucom a t • u s eyes (Fig. 3). For comparison, an R O C curve for the structural parameter disc rim area is shown in Fig. 4 for the same database. The curve indicates a better test than either standardized or non-standardized pallor measurements• Further analysis revealed that the pallor density dis-
435
tributions were similar for the normal subjects and the glaucomatous patients. Figure 5 is a summary of the frequency distributions and shows that neither the position of the two curves or the distribution widths were significantly different between the two groups. Discussion
Attempts to quantify optic disc pallor date to 1950, when Boch compared the color of the disc to different shades of pink paints [2]. Gloster in 1967 developed a technique to estimate the oxygen saturation of optic nerve blood by reflectometry [6]. Berkowitz and Balter, in 1970, quantified optic disc pallor by measuring the amount of yellow or magenta filtered light required to match the color of the optic disc viewed simultaneously through a modified ophthalmoscope [1]. Davies, in 1970, used a photographic method in which he took color photos of the disc, illuminated the photos under red and green light and developed a model for the relative blood layer thickness of the optic nerve head [4]. Schwartz et al. developed a method for quantifying the area of optic disc pallor [17]. Schwartz and Kern attempted microdensitometry but had difficulty standardizing the measurements from image to image [16]. Results of pallor measurements with an electron scanner have been reported by Rosenthal et al. in 1980 [14]. To minimize the variability inherent with photographic techniques a boundary method was developed by Nagin et al. in 1985 [10]. The reproducibility of pallor measurements made in this way was 2%-8%. It is well accepted that optic disc pallor progresses with glaucomatous damage. This would seem to occur because additional white lamina cribrosa is uncovered to the viewer as the vascularized neural tissue is lost. In retrospective studies of longitudinal data, Schwartz demonstrated an increase of pallor in ocular hypertensives over time [9, 15]. Robert and Maurer evaluated the effect of artificially elevated IOP on optic disc pallor in normal and glaucomatous eyes and found an increase in pallor with decreasing perfusion pressure [13]. Changes in optic disc pallor have been reported to occur before defects of the visual field have been detected [11, 18]. Stereoscopic optic nerve head photographs have been used for some time in the routine clinical care of glaucoma patients. They represent the best currently available method to carefully monitor the appearance of the optic nerve head over time, but their subjective and qualitative evaluation has prompted a search for more quantitative and objective techniques. Newer techniques of digital video imaging avoid the problems with variation of film, its exposure, and development. We used computerized videographic image analysis to obtain reflective measurements of the optic disc surface. Mikelberg et al. have also reported on the reproducibility of computerized pallor measurements obtained with the Rodenstock instrument and found that the coefficients of variation ranged from 8.3%-20.1% for intervideographic variability of different pallor histogram
values [7]. A significant difficulty with repeated pallor measurements of this type has been that the frequency distribution of pallor values would shift to the left and to the right across pallor values, though the shape of the distribution changed little. Miller and Caprioli found that the reproducibility of repeated pallor measurements was substantially reduced by measuring the width of the distribution rather than its absolute position, thereby standardizing the measurements to the redness of the disc vessels [8]. Variability of pallor measurements made in this way was least for the difference between trimmed mean parameters h2 and h5 (3.4%) and in a small group of subjects, the pallor frequency distribution of the glaucomatous patients was found to be wider than in normal subjects. The distribution width, h2-h5, is a measure of image contrast between red vessels and portions of the disc with the greatest pallor. Using the distribution width as opposed to absolute pallor measurements provided a measure that was not sensitive to the variable shifts in position of the frequency distributions. The lack of a videographically detectable pallor difference between the normal and glaucomatous groups here precluded the useful differentiation of early glaucoma from normal with these methods. The database was chosen to provide a rigorous test for a discriminating parameter by including patients with early damage and excluding patients with advanced glaucomatous damage. Digital image analysis of pallor derived from the relative monochromatic reflectances of the optic disc do not perform as well as structural parameters to differentiate between normal and glaucomatous eyes. The comparison made here with rim area is based on the same dataset since only internal comparisons can be made to test the relative values of different parameters. The relative values of different quantitative optic nerve head measurements with which to detect progressive changes longitudinally remains an important unanswered question which is presently under study.
References 1. Berkowitz JS, Baiter S (1970) Colorimetric measurements of the optic disc. Am J Ophthalmol 69:385-386 2. Boch RH (1950) Zur klinischen Messung der Papillenfarbe. Ophthalomologica 120:174-177 3. Caprioli J, Miller JM (1988) Videographic measurements of optic nerve topography in glaucoma. Invest Ophthalmol Vis Sci 29:1294-1298 4. Davies EG (1970) Quantitative assessment of colour of the optic disc by a photographic method. Exp Eye Res 9:106-113 5. Flammer J, Drance SM, Augustiny L, Frankhauser A (1985) Quantification of visual field defects with automated perimetry. Invest Ophthalmol Vis Sci 26:176-181 6. Gloster J (1967) Fundus oximetry. Exp Eye Res 6:187-212 7. Mikelberg FS, Douglas GR, Drance SM, Schulzer M, Wijsman K (1988) Reproducibility of computerized pallor measurements obtained with the Rodenstock Disk Analyzer. Graefe's Arch Clin Exo Ophthalmol 226:269-272 8. Miller J, Caprioli J (1988) Videographic quantification of optic disc pallor. Invest Ophthalmol Vis Sci 29 : 320-323 9. Nagin P, Schwartz B (1985) Detection of increased pallor over time. Ophthalmology 92:252-261
436 10. Nagin P, Schwartz B, Nanba K (1985) The reproducibility of computerized boundary analysis for measuring optic disc pallor in the normal optic disc. Ophthalmology 92:243-251 11. Pederson JE, Anderson DR (1980) The mode of progressive disc cupping in ocular hypertension and glaucoma. Arch Ophthalmol 98 : 490 12. Prescott P, Hogg RV (1977) Trimmed and outer means and their variances. Am Stat 31 : 156-157 13. Robert Y, Maurer W (1984) Pallor of the optic disc in glaucoma patients with artificial hypertension. Doc Ophthalmol 57:203214
14. Rosenthal AR, Falconer DG, Barret P (1980) Digital measurement of pallor-disc ratio. Arch Ophthalmol 98:2027-2031 15. Schwartz B (1980) Optic disc changes in ocular hypertension. Surv Ophthalmol 25 : 148-154 16. Schwartz B, Kern J (1977) Scanning microdensitometry of optic disc pallor in glaucoma. Arch Ophthalmol 95:2159-2165 17. Schwartz B, Reinstein NM, Lieberman DM (1973) Pallor of the optic disc. Arch Ophthalmol 89:278-286 18. Sommer A, Pollack I, Maumenee AE (1979) Optic disc parameters and onset of glaucomatous visual field loss. I. Methods and progressive changes in disc morphology. Arch Ophthalmol 97:1444-1448
Graefe's Arch Clin Exp Ophthalmol (1992)230:432-436
for Cgnical and Experimental
Ophthalmology © Springer-Verlag 1992
Digital image analysis of optic nerve head pallor as a diagnostic test for early glaucoma* Albert Assad and Joseph Caprioli Glaucoma Section, Yale University School of Medicine, Department of Ophthalmology and Visual Science, 330 Cedar Street, New Haven, CT 06510, USA Received October 14, 1991 /Accepted December 23, 1991
Abstract. We developed a computer-based technique to quantify optic nerve head pallor from videographically acquired digitized optic nerve images and tested the ability of pallor measurements to discriminate between normal eyes and eyes with early glaucoma. Corresponding pixel values from images obtained under 540 nm (green) and 640 nm (red) light with a videographic fundus camera were used to quantify optic nerve head pallor. A pallor density histogram was calculated for each eye, and contained values betwen 0 (red) to 1 (white). A measure o f the distribution width o f the histogram provided pallor measurements standardized to the measurements o f the large veins o f the disc. A database o f one eye each of 44 normal controls and 70 patients with early open angle glaucoma was used to test the measurements for diagnostic sensitivity and specificity. These standardized pallor measurements did not perform better than absolute pallor measurements to discriminate between normal and glaucomatous eyes. The sensitivity and specificity of standardized pallor measurements (49% and 57%, respectively, for this database) were not as good as those for stereoscopic measurements o f disc rim area in the same database (70% and 73%). Pallor measurements of this type do not appear to be sensitive or specific indicators of early glaucoma.
Introduction The poor ability o f intraocular pressure measurements to differentiate between normal and glaucomatous eyes and the subjectivity and variability of visual field measurements have placed additional emphasis on objective measurements of physical attributes of the optic disc * Supported in part by grants from Research to Prevent Blindness, Inc, The Connecticut Lions Eye Research Foundation, and Alcon Laboratories, Inc. ; the authors have no proprietary interest in the instruments or softwa/e described within
Correspondence to: J. Caprioli
and nerve fiber layer to detect early glaucoma. Decisions about diagnosis or treatment are often based on a change in the appearance o f the optic disc during the course o f the disease. Pallor o f the optic disc is considered an important sign o f optic nerve atrophy and is a manifestation of the loss o f vascularized neural tissue. This study was performed to evaluate the ability of pallor measurements derived from digital videographic imaging to differentiate between normal eyes and those with early glaucoma.
Materials and methods
Subjects A database was constructed from images of 44 normal subjects and 70 glaucoma patients. The glaucoma patients were agematched to the normal subjects. Normal subjects were recruited from hospital staff and friends or spouses of patients; none had presented for medical evaluation. Patients were considered glaucomatous if they had elevated intraocular pressure or a history of elevated intraocular pressure before treatment, and early to moderate, typical glaucomatous visual field defects. Typical glaucomatous visual field defects were defined, in a reliably performed automated threshold visual field test, as at least: 1. Two or more contiguous points with 10 dB loss or greater in the superior or inferior Bjerrum areas, compared with perimeterdefined age-matched controls 2. Three or more contiguous points with 5 dB loss or greater in the superior or inferior Bjerrum areas 3. A 10 dB difference across the nasal horizontal midline in two or more adjacent locations. The most superior and inferior rows of threshold measurements from the 30° programs were excluded from these criteria to avoid the inclusion of rim artifacts. Visual field indices mean defect, corrected loss variance, and short term fluctuation as described by Flammer et al. were calculated for each visual field [5].
Image analysis Image acquisition of the optic disc was performed with a videographic instrument designed for this purpose (Rodenstock Instru-
433 ments, Munich, FRG). We developed software, which ran on a stand-alone PC to construct pallor density histograms from the digitized images. The reproducibility of these measurements has previously been reported [8]. The digitized images stored by the videographic system consist of a matrix of 256 x 256 points. Each pixel has a value eight bits deep; values ranged from 0-255. The corresponding pixel values from green and red monochromatic images obtained with a videographic fundus camera were used to calculate the pallor of the optic nerve head according to equation 1. The green digitized image was obtained under 540 nm light and the red was obtained under 640 nm light. The pallor density of each pixel was defined as twice the green reflectance divided by the sum of the green and red reflectance: 2 x (green reflectance) pallor density - (red reflectance) + (green reflectance)
(1)
The selection of Eq. (1) for calculating pallor was based on the model described by Davies [4]. In this model, a grey-white nerve head, which is equally reflective to both red and green light, is covered by a thin layer of blood. The relative blood layer thickness (RBLT) is proportional to the difference of the log of the reflectances at green and red wavelength: RBLT = log (green reflectance) - log (red reflectance) + 1
To compare different test parameters, receiver-operator characteristic (ROC) curves were used. ROC curves display the entire range of sensitivity and specificity values and provide a means for evaluating whether a certain test or parameter is better than another at detecting differences between groups. The greater the area under the curve, the better the test is at discriminating between the normal and glaucomatous groups.
Results
Summary data for age, sex, refractive error, visual field indices, and rim area of the optic nerve head are given
Table 1. Descriptive statistics for age, sex, refractive error, visual field, and optic nerve head* Normal
Glaucoma
Males Females
14 30
35 35
Age (years)
59.48 -+ 11.56
60.04_+ 10.53
0.73_+ 1.70
-1.09_+ 3.18
-0.62_+ 1.62 2.20_+ 2.87
7.11_+ 5.79 35.89_+36.31
1.07-t- 0.20
0.82_+ 0.30
(2)
A value of 1 is added to provide an appropriate range of numbers : these range from 0 for a thick blood layer to 1 for the presence of no blood. Equation (1) is an approximation good to within 4% of Eq. (2) for pallor values
0.4
CC
hi
h6
O II.
Ill 15% I
°.
I PALLOR
RED
15%
30%
I
0.2
I
k"
VALUE WHITE
Fig. 1. This figure shows the technique of using trimmed means for describing non-Gaussian distributions. HI is the average pallor value for the whitest 5% of the points and h6 is the average pallor value for the reddest 5% of the points. H2-h5 is the difference between the trimmed means h2 and h5
0.0
:
0.0
I
0.2
FALSE POSITIVE
I
I
0.4 RATE
0.6
I
0.8
1.0
(1-specificity)
Fig. 2. This figure shows the ROC curve for the trimmed mean hl ; hl represents an absolute pallor value
434
1.0
1,0
I
|
=
"~ 0 . 8 c o
__l:
. . . . . .
I
__
0.8
Ul 1-.
0.4
1--
,•
U) 0
I
0.6
0.6 MI
,.
i
0.2
0.2
..
•
•
00o
w er
k-
I-
0.0
r0.0
,
,
0.2
j
0.4
F A L S E P O S I T I V E RATE
0.0
,
0.6
0.8
1.0
I
I
I
, 0.2
J 0.4
FALSE P O S I T I V E RATE
(1-specificity)
Fig, 3. This figure shows an ROC curve for the difference between trimmed mean parameters h2 and h5. H2-h5 is a measure of the width of the pallor frequency distribution
60
0.0
j 0.6
I
0.8
1.0
(1 -specificity)
Fig. 4. An ROC curve for the structural parameter rim area is shown for comparison. This parameter more efficienty discriminated between normals and glaucomas than did the pallor measurements
I
50
lid
r3 < :f i \ v)
,.J ILl uX L
40
30
L
•
'~
/
20
#
,
to
It
t
Fig. 5. The average pallor value distributions for the n o r m a l subjects and glaucoma patients. There is a slight shift t o w a r d increased pallor values in the glaucoma patients, but this was n o t a statistically significant difference. - - N o r m a l Subjects . . - G l a u c o m a Patients
L
10
0
I
0.0
red
0.2
i
i
0.4
0.6
PALLOR
I
0.8
1 .(
white
in Table 1. There was no significant difference between the mean ages of the two groups. Figure 2 is an R O C curve for the non-standardized test parameter hi. H 1 represents the average pallor value of the most pale 5% of the image pixels. This parameter was unable to differentiate well between normal subjects and glaucomatous patients as evidenced by the relatively diagonal line in the R O C curve. The difference between trimmed mean parameters h2 and h5 reflects the distri-
bution width, and is independent of the position of the distribution. This standardization did not improve the ability to discriminate between these normal and glaucom a t • u s eyes (Fig. 3). For comparison, an R O C curve for the structural parameter disc rim area is shown in Fig. 4 for the same database. The curve indicates a better test than either standardized or non-standardized pallor measurements• Further analysis revealed that the pallor density dis-
435
tributions were similar for the normal subjects and the glaucomatous patients. Figure 5 is a summary of the frequency distributions and shows that neither the position of the two curves or the distribution widths were significantly different between the two groups. Discussion
Attempts to quantify optic disc pallor date to 1950, when Boch compared the color of the disc to different shades of pink paints [2]. Gloster in 1967 developed a technique to estimate the oxygen saturation of optic nerve blood by reflectometry [6]. Berkowitz and Balter, in 1970, quantified optic disc pallor by measuring the amount of yellow or magenta filtered light required to match the color of the optic disc viewed simultaneously through a modified ophthalmoscope [1]. Davies, in 1970, used a photographic method in which he took color photos of the disc, illuminated the photos under red and green light and developed a model for the relative blood layer thickness of the optic nerve head [4]. Schwartz et al. developed a method for quantifying the area of optic disc pallor [17]. Schwartz and Kern attempted microdensitometry but had difficulty standardizing the measurements from image to image [16]. Results of pallor measurements with an electron scanner have been reported by Rosenthal et al. in 1980 [14]. To minimize the variability inherent with photographic techniques a boundary method was developed by Nagin et al. in 1985 [10]. The reproducibility of pallor measurements made in this way was 2%-8%. It is well accepted that optic disc pallor progresses with glaucomatous damage. This would seem to occur because additional white lamina cribrosa is uncovered to the viewer as the vascularized neural tissue is lost. In retrospective studies of longitudinal data, Schwartz demonstrated an increase of pallor in ocular hypertensives over time [9, 15]. Robert and Maurer evaluated the effect of artificially elevated IOP on optic disc pallor in normal and glaucomatous eyes and found an increase in pallor with decreasing perfusion pressure [13]. Changes in optic disc pallor have been reported to occur before defects of the visual field have been detected [11, 18]. Stereoscopic optic nerve head photographs have been used for some time in the routine clinical care of glaucoma patients. They represent the best currently available method to carefully monitor the appearance of the optic nerve head over time, but their subjective and qualitative evaluation has prompted a search for more quantitative and objective techniques. Newer techniques of digital video imaging avoid the problems with variation of film, its exposure, and development. We used computerized videographic image analysis to obtain reflective measurements of the optic disc surface. Mikelberg et al. have also reported on the reproducibility of computerized pallor measurements obtained with the Rodenstock instrument and found that the coefficients of variation ranged from 8.3%-20.1% for intervideographic variability of different pallor histogram
values [7]. A significant difficulty with repeated pallor measurements of this type has been that the frequency distribution of pallor values would shift to the left and to the right across pallor values, though the shape of the distribution changed little. Miller and Caprioli found that the reproducibility of repeated pallor measurements was substantially reduced by measuring the width of the distribution rather than its absolute position, thereby standardizing the measurements to the redness of the disc vessels [8]. Variability of pallor measurements made in this way was least for the difference between trimmed mean parameters h2 and h5 (3.4%) and in a small group of subjects, the pallor frequency distribution of the glaucomatous patients was found to be wider than in normal subjects. The distribution width, h2-h5, is a measure of image contrast between red vessels and portions of the disc with the greatest pallor. Using the distribution width as opposed to absolute pallor measurements provided a measure that was not sensitive to the variable shifts in position of the frequency distributions. The lack of a videographically detectable pallor difference between the normal and glaucomatous groups here precluded the useful differentiation of early glaucoma from normal with these methods. The database was chosen to provide a rigorous test for a discriminating parameter by including patients with early damage and excluding patients with advanced glaucomatous damage. Digital image analysis of pallor derived from the relative monochromatic reflectances of the optic disc do not perform as well as structural parameters to differentiate between normal and glaucomatous eyes. The comparison made here with rim area is based on the same dataset since only internal comparisons can be made to test the relative values of different parameters. The relative values of different quantitative optic nerve head measurements with which to detect progressive changes longitudinally remains an important unanswered question which is presently under study.
References 1. Berkowitz JS, Baiter S (1970) Colorimetric measurements of the optic disc. Am J Ophthalmol 69:385-386 2. Boch RH (1950) Zur klinischen Messung der Papillenfarbe. Ophthalomologica 120:174-177 3. Caprioli J, Miller JM (1988) Videographic measurements of optic nerve topography in glaucoma. Invest Ophthalmol Vis Sci 29:1294-1298 4. Davies EG (1970) Quantitative assessment of colour of the optic disc by a photographic method. Exp Eye Res 9:106-113 5. Flammer J, Drance SM, Augustiny L, Frankhauser A (1985) Quantification of visual field defects with automated perimetry. Invest Ophthalmol Vis Sci 26:176-181 6. Gloster J (1967) Fundus oximetry. Exp Eye Res 6:187-212 7. Mikelberg FS, Douglas GR, Drance SM, Schulzer M, Wijsman K (1988) Reproducibility of computerized pallor measurements obtained with the Rodenstock Disk Analyzer. Graefe's Arch Clin Exo Ophthalmol 226:269-272 8. Miller J, Caprioli J (1988) Videographic quantification of optic disc pallor. Invest Ophthalmol Vis Sci 29 : 320-323 9. Nagin P, Schwartz B (1985) Detection of increased pallor over time. Ophthalmology 92:252-261
436 10. Nagin P, Schwartz B, Nanba K (1985) The reproducibility of computerized boundary analysis for measuring optic disc pallor in the normal optic disc. Ophthalmology 92:243-251 11. Pederson JE, Anderson DR (1980) The mode of progressive disc cupping in ocular hypertension and glaucoma. Arch Ophthalmol 98 : 490 12. Prescott P, Hogg RV (1977) Trimmed and outer means and their variances. Am Stat 31 : 156-157 13. Robert Y, Maurer W (1984) Pallor of the optic disc in glaucoma patients with artificial hypertension. Doc Ophthalmol 57:203214
14. Rosenthal AR, Falconer DG, Barret P (1980) Digital measurement of pallor-disc ratio. Arch Ophthalmol 98:2027-2031 15. Schwartz B (1980) Optic disc changes in ocular hypertension. Surv Ophthalmol 25 : 148-154 16. Schwartz B, Kern J (1977) Scanning microdensitometry of optic disc pallor in glaucoma. Arch Ophthalmol 95:2159-2165 17. Schwartz B, Reinstein NM, Lieberman DM (1973) Pallor of the optic disc. Arch Ophthalmol 89:278-286 18. Sommer A, Pollack I, Maumenee AE (1979) Optic disc parameters and onset of glaucomatous visual field loss. I. Methods and progressive changes in disc morphology. Arch Ophthalmol 97:1444-1448
