- Email: [email protected]
A new computer assisted objective method for quantifying vascular changes of the bulbar conjunctivae
Copyright
Ophthal. Physiol. Opt. Vol. 16, No. 5, pp. 430-437, 1996 0 1996 The College of Optometrists. Published by Elsevier Science Ltd Printed in Great Britain 0275-5408196 $15.00 + 0.00
PII: SO2755408(96)00037-3
A new computer assisted objective method for quantifying vascular changes of the bulbar conjunctivae Christopher
G. Owen,’
’ Applied Goswell London,
Research Centre, City University, Dame Alice Owen London ECIV 7DD, UK; ‘Institute of Ophthalmology, Bath Street, London ECIV 9EL, UK
Vision Road, 11-43
Fred W. Fitzke2
and E. Geoffrey
Woodward’ Building, 31 I-321 University College
Summary A novel computer software method was used to quantify the conjunctival plexus on the scleral background for measurement of the vascular surface area from photographs. A previously described method was used (Palmer, J. R., Owen, C. G., Ford, A. M., Jacobson, R. E. and Woodward, E. G. (1996). Optimal photographic imaging of the bulbar conjunctival vasculature. Ophthal. Physiol. Opt. 16, 144- 149) to optimise photographic imaging of the bulbar conjunctival vasculature by increasing the information content in the image. Repeatability of this technique was evaluated. Twenty subjects (20 eyes) free from ophthalmological and systemic abnormality were examined on two separate occasions. The maximum 95% confidence limits for repeatability are + 8.58/ - 3.95%. For 10 consecutive estimates of vascularity the maximum 95% confidence interval lie between +6.54%. To evaluate the technique the lateral-bulbar conjunctivaein 10 soft (SCL) and 10 rigid gas permeable contact lens (RGPCL) wearers during the first 10 months of contact lens wear, were assessed and compared with subjective grading of hyperaemia. The new method showed sufficient sensitivity in detecting increased hyperaemia in the RGPCL wearing group and demonstrated statistically significant change. Subjective graded assessment of vascularity (using established classifications) detected increased hyperaemia, however, this was not statistically significant. Conjunctival vasculature is a dynamic structure and a source of valuable quantitative information where the ocular environment is varied, or where the ocular surface is affected by disease. Hence it is worthy of further investigation. A simple inexpensive method of computer assisted determination of vascularity is described. Copyright 0 1996 The College of Optometrists. Published by Elsevier Science Ltd.
(1989) found that existing edge detection algorithms for detecting retinal vasculature were unsatisfactory. The grey level profile of a cross section of a blood vessel was approximated by a Gaussian shaped curve. A total of 12 different Gaussian matched filters were devised and used to detect vessel segments. Tamura et al. (1988) used a similar technique to detect retinal vasculature with a spatial filter derived from a second order Gaussian filter. Vasculature detection has also been conducted in the cochlea to determine whether changes in blood flow may be related to hearing loss (Miles and Nuttall, 1993). The conjunctival plexus consists of a multi-layered intricate vascular pattern which is readily observed through its transparent mucus membrane. Its vascular supply to the perilimbal region (4 to 6 mm from the limbus) of the bulbar conjunctivae originates from the ophthalmic artery and
Introduction Examination of vasculature has been considered as a useful source of information for many disorders. Numerous sites have been studied where vasculature can be readily observed. Increasingly, computational methods have been devised to quantify vasculature. Automated approaches have been developed to detect retinal vasculature (Chaudhuri et d., 1989). In 1992 Spencer et al. quantified retinal microaneurysms using digital image processing. Chaudhuri et al. Correspondence and reprints requests ro: Mr. C. G. Owen, Research Fellow, Applied Vision Research Centre, Department of Optometry and Visual Science, Dame Alice Owen Building, 311-321 Goswell Road, London EClV 7DD, UK. Received: 27 March 1996 Revisedfom: 8 May 1996
430
Quantifying
traveis to the limbus via the anterior cilary artery. Systemic disorders such as diabetes (Ditzel and Saglid, 1954; Coget et al., 1989), hypertension (Davydova et al., 1990), and sickle cell disease (Paton, 1961; Sejeant et al., 1972) can affect the morphological and morphometric characteristics of the conjunctival vasculature. Mo~hologica~ alterations in the human bulbar conjunctivae in diabetes are said to include an increase in surface vascularity compared to normals (Chen et al., 19X6), as opposed to hypertensive changes which have diametrically opposite trends, including a decrease in vascularity, length, diameter, large vessels and terminal vessels, compared to normals (Chen et ai., 1986). Such findings may be confounded by vasculature changes associated with age (McMonnies and Ho, 1991; Kovalchek et al., 1984). Ocular inflammation, cornea1 hypoxia, and ocular xerosis may also affect the hyperaemic state of the bulbar conjunctivae. According to McMo~ies and ~apman-Davies (1987) conjunctival hyperaemia occurs ‘commonly’ in contact lens wearers. Chronic conjunctival hyperaemia develops in response to prolonged toxic or physiological insult, or surface deposits or other forms of lens degradation. However, Dixon and Lawaczeck (1963) report that frequent ‘overfilling’ of limbal vessels is common place with asymptomatic hard contact lens wear. Ruben et al. (1976) report increased bulbar conjunctivae hyperaemia with asym,ptomatic soft contact lens wear. McMonnies et al. (1982) conclude that soft contact lens wear induces chronic inferior limbal injection which is not seen with hard contact lens wear. Current techniques to assess hyperaemia of the bulbar conjunctivae, principally rely on subjective assessment, with judgements made from inspection of graded photographic plates (McMonnies and Chapman-Davies, 1987). Such ordinal ranking systems will only allow measurements to be made within certain ‘bracketed’ amounts, thus reducing its sensitivity. Such methods may be open to subjective bias as well as inter subject variabilities such as pupil size, iris colour, palpebral aperture size and skin colour. Investigation of the effect of these physiological variations on the subjective gradings of hyperaemia has never been investigated. A number of computer assisted techniques of conjunctival hyperaemic assessment have been described (Chen et al., 1987; Villumsen et al., 1991; Willingham et al., 1994, 1995a, 1995b). Chen et al. (1987) used a Nikon F3 35 mm camera with macro lens giving a 4.5 x magnification in the film plane. Black and white 35mm film images (ASA 50) containing 6.7 mm2 of conjunctival area were digitised to a 128 x 128 pixel image. This was averaged using a 5 x 5 pixel square and any background luminosity gradients were corrected. The enhanced image was subtracted from the original image, averaged again, and thresholded to isolate vasculature. Villumsen et al. (1991) stored colour transparencies of the temporal bulbar conjunctivae on to a video camera. Video signals were then converted into a mono-
vasculature
changes:
C. G. Owen
et al.
431
chrome 270 x 300 pixel image. A 3 x 3 pixel square was then averaged and a simple edge detection filter applied. The resultant image was then thresholded using a manually selected pixel intensity of a vessel region. This technique was devised to avoid a laborious, time consuming manual point counting system (Vill~sen and Alm, 1989) employed to monitor the conjunctival vascular response to topical application of prostraglandin F,,-isopropylester eye drops. Willingham et al. (1995b) described an automated qualitative measurement of conjunctival hyperaemia using colour images. Images were obtained using a Nikon FS-2 photoslit lamp and recorded to a video image, with a region of interest (ROI) of 3.8 cm*. Using dedicated software the images, stored on magnetic disc (640 pixels x 480 pixels), were automatically masked to exclude specular reflections, lashes, lids and un-illuminated regions. The red (R), green (G), and blue (B) colour content of each pixel was extracted. For each colour the intensity ranged from 0 (darkest) to 255 (brightest). The total intensity (I) of a pixel was the sum of the RGB components. Mean relative redness (RR) was calculated for each pixel from the ratio of the intensity of R to I. The measure of RR was the sum of these ratios divided by the number of active pixels multiplied by 100. To compute vessel area (VA), red free images (G) were selected because of superior vessel contrast. Five Gaussian spatial filters (SD 1, d2, 2, 242, 8 pixels) were applied. Blurring of images allowed correction of luminance gradients over the active pixel region. Vessel and non-vessel areas were rn~u~ly labelled. Typical pixel patterns in feature space spanning a vessel border were modelled by multivariate Gaussian densities. This model was then used to isolate vasculature over the ROT. Once a binary image is obtained morphological and morphometric characteristics can be obtained. For the purposes of vascular examination with contact lens wear it is vessel area which is of most importance, assuming vascular morphology remains unchanged. We describe a simple and inexpensive technique of objectively quantifying vessel area from monochrome images which contain high information content. Unlike previous methodologies of automated vascular examination analysis, repeatabili~ measurements of the current technique were assessed. This technique was used to assess the vascular response of the lateral bulbar conjunctivae in asymptomatic soft contact lens and rigid gas permeable contact lens wear. Palmer et al. (1996) have investigated the optimum colour filter required for imaging the conjunctival vasculature using the Nikon FS-2 photo-slit lamp and Ilford HP5 film. Optimum filtration enables images of superior information content to be obtained. This increases the quality of the binary image generated by the computer programme (Owen et al. f 1995a). If any component of the system were to change, e.g. the use of a charged couple device (CCD) detector instead of HP5 film, the optimum filter would need to be re-calculated.
432
Materials
Ophrhal.
Physiol.
Opt. 1996 16: No 5
and methods
The experimental set up described by Palmer et al. (1996) was used in the current study. Images of the dextro temporal bulbar conjunctivae were captured using a Nikon FS-2 photo-slit lamp with a xenon flash tube, on Ilford HP5 monochrome film. The angle between observation and illumination systems was set to 30”, with the optical axis being approximately normal to the curved conjunctival surface. This has been experimentally found to minimise specular reflections from the conjunctival epithelium (Palmer et al., 1996). Photographic images were scanned using a Polaroid SprintScan 35mm Slide Scanner and saved as a tagged image format file (TIFF). A square region of interest (ROI) immediately lateral to the limbus of 17mm* was manually selected tangential to the limbal conjunctival border. All images were scanned in an identical manner to a 336 x 336 pixel square to a resolution of 133 dots per inch. The 8 bit, 256 grey level image was spatially averaged fifteen times consecutively (using a 3 x 3 pixel square on each occasion), and then subtracted from the original image spatially averaged once. Spatially averaging in a repetitive sequence (X 15) will smooth all high contrast gradients, such as vasculature, leaving low contrast gradients of the background relatively unchanged. When this is subtracted from an image averaged only once, the high contrast gradients (vasculature) remain, Contrast enhancement was performed, which merely allows improved visualisation of the images but does not affect the raw data used for analysis. This was experimentally achieved by multiplying the resultant by 4, and by adding a constant to each pixel grey level using Lucida version 2 software. Figures 1 (a) and (6) show the image of the conjunctival vasculature pre and post processing. Once the image is processed, a frequency distribution of the grey level of each pixel in the image is plotted (see Figure 2). This plot represents the addition of the scleral and conjunctival vasculature grey level profiles. Threshold selection is of importance as this determines the degree of vascular representation. Thresholds were taken at the peak, 10% below the peak grey level, and 20% below the peak (a)
lb)
py’ 0
/ 50
100
Grey level (O-256) Figure 2. Hypothetical frequency distribution plot of each pixel grey level in the image. Thresholds for the binary image are taken at the peak, 10% below, and 20% below as indicated with vertical solid lines right to left respectively. The frequency distribution represents the combined scleral and vasculature grey level profiles. Thick solid line = resultant grey level profile, dashed line = hypothetical scleral content, and dotted line = hypothetical vasculature content.
grey level, as the cut-off between vasculature and sclera to create the binary image (see Figure 3). Both the episcleral and conjunctival plexus will be represented in differing amounts. Selection of the most appropriate threshold is given in the results section. Once a binary image is achieved the percentage of vasculature within the 17 mm2 area is calculated. This is equivalent to a dimensional value in terms of the percentage area of the ROI (17 mm’) which is considered as vasculature. Repeatability Twenty subjects (20 right eyes) free from ophthalmological and systemic abnormalities were photographed on 2 separate occasions to evaluate repeatability. Age ranged from 30 to 53 years (mean 41.5 years). All subjects were non contact lens wearers and were seen at the same time of day for each examination (k 1% hours) approximately 3 weeks apart. On arrival a questionnaire was completed, symptoms and history recorded. This allowed sufficient time to adapt to the ambient environment. This was followed by slit-lamp photography of the dextro-lateral bulbar conjunctiva. A full anterior eye examination was carried out including subjective grading of hyperaemia and fluorescein staining. Statistical analysis
Figure 1. (a) and (b): Images of the lateral-bulbar junctivae pre and post image processing, respectively.
con
Proportion of vasculature in the image (expressed as a percentage) was calculated at (a) the peak of the grey level
Quantifying
vasculature
changes:
C. G. Owen
et al.
433
(b)
(a)
Figure 3. Shows a binary image obtained (b) 10%
below
the peak
grey
level,
for threshold taken at (a) the peak of the grey level histogram, and (c) 20% below the peak grey level.
plot, ;b) 10% below, and (c) 20% below. These were analysed separately comparing sessions 1 and 2. (Graphs were drawn for pairs of data measured at each level (a, b and c), Correlation coefficients were evaluated as a measure of the association between pairs of data. Proximity of the points to the regression line gives some indication of the agreement. An association where the regression line intercepts with the origin, and has a gradient of one, was thought to offer the most suitable threshold. A preferred alternative to the correlation approach is to consider the difference between paired values obtained at session 1 and 2, plotted against their mean for each vascularity value (Rudnicka et al., 1992). These plots allow investigation of any relationship between the measurement error and true value. When the true value is unknown a mean of the measurements is the best estimate (Bland and Altman, 1986). The coefficient of repeatability can be calculated, defined by the British Standards Institute (BS5497) to be twice the standard deviation (SD) from the mean difference. By definition, 95 % of the values will lie between +2 SD. Provided that the differences within *2 SD would not be clinically important one can say there is good repeatability between the two measures. Ideally the threshold selected should have the smallest confidence range. Conjunctival vasculature is a dynamic structure, hence 10 consecutive photographs of the same eye were taken. The rationale being that there would be minimal vasculature change over the short period of photography. The mean f2 SD was calculated to give another measure of variance in the estimated percentage of vasculature. This gives an indication of intra-session variabilities. Objective vs subjective assessment of hyperaemia Vascularity of the dextro-lateral-bulbar conjunctiva in 10 soft (Bausch and Lomb 70) and 10 rigid (Bausch and Lomb Quantum) daily contact lens wearers was evaluated by this novel technique, and compared with subjective grading. Measurements were taken pre contact lens fitting, after
1 month, 4 months, and 10 months of contact lens wear. Patients were all members of the UK fire service. All aspects of both the bulbar and palpebral conjunctivae were examined and graded. Vascular&y of the bulbar conjunctivae was subjectively classified from 1 to 5 where 1 represents normal vascularity, and 5, gross hyperaemia and extreme vessel dilation. This classification was based on the method described by McMonnies and Chapman-Davies (1987) with reference to photographic plates. Although these showed the inferior aspect of the bulbar conjunctivae, the vascular appearance was extrapolated to the lateral portion. As changes in normal contact lens wear are likely to be small, a grading of 1 to 5 was thought to be sufficient. In addition to the subjective measurement, observed over the entirety of the lateral conjunctivae, a masked area (17 mm’) immediately adjacent to the limbus was objectively quantified, by the method described above. Direct comparison of the 2 methods, although useful, should be treated cautiously as differing areas of vascularity were assessed. Results Repeatability Data obtained at session 2 was plotted against session 1, for 15 eyes, measured at (a) the peak value, (b) 10% below, and (c) 20% below the peak. Five eyes were rejected due to an experimental oversight. Correlation analysis followed by linear regression by the least squares method was performed. Values for R2 and equation of the regression lines are given in Table 1. Figure 4 shows the difference between paired values obtained at session 1 and 2, plotted against their mean hyperaemic value as measured at each threshold. This allows a more valid assessment of agreement. Individual data points show an acceptable spread about the mean at all x values, with no apparent ‘funnelling effect’. The 95 % confidence limits for inter-session repeatability are +9.85/ -8.51% (spread 18.36%) when measured at the peak of
434
Ophthal.
Physiol.
Opt.
1996
16: No 5
the grey level distribution plot, +8.58/-3.95% (spread 12.53%) at 10% below the peak, and +6.26/-5.98% (spread 12.24%) at 20% below. Figure 5 shows 10 consecutive estimates of vascularity, as measured at each threshold level. The mean value is indicated in each case by the dashed line, and the solid lines represent f 2 SD from the mean. The 95 % confidence
Table 1. Values for R2 and equation of the regression line measured at the peak threshold, 10% below, and 20% below Measuring
level
Equation
of regression
y = 0.7976x y = 1.0248x y = 0.6766x
Peak threshold 10% Below peak 20% Below peak
R2 value
line
0.4221 0.757 0.5393
+ 12.95 + 1.6664 + 3.9854
(a)
50
lb)
60
55
Mean vascularity
percentage
65
70
75
as measured at the peak histogram level (%),
10 A
T@
8--
b s
6--
25
4--
% E 3 2 1
2
-2 --
.c
-4
8 ii
-6 --
z g
d
* d
A d
A d
A
o--
A
A
a
A A
A
-8 -I
-10 7 20
15
10
Mean vascularity
percentage
35
30
25
40
as measured at 10% below the peak (%).
(c’ &‘j 0 0
t .:
2 --
&
o-
0 0
0 00
a !i .E ' Es $ g
0
-2 -4
-6-
0
0
0
-8 -10 7 0
I 2
4
Mean vascularity
6
8
percentage
10
12
14
16
18
20
as measured at 20% below the peak (%).
Figure 4. Differences between paired values of vascularity for sessions 1 and 2 plotted against their mean (n = 15). Thick solid line represents mean difference and the thin solid line 95% confidence limits, where (a) is measured at the peak, (b) 10% below the peak, and (c) 20% below the peak.
Quantifying
; .z
70
‘g;
60 50
0
vasculature
0
0
5
6
changes:
C. G. Owen
et al.
435
0
D
0 ‘ij I k 8 * z 0 -e
40 30 20 10 0 0
1
3
2 Number
4 of consecutive
estimates
Figure 5. Ten consecutive estimates of vascularity (open circles); (b) 10% below peak (open triangles), oeak (ooen sauares). The dashed line represents solid line +-2SD.
limits are * 10.93 % when measured at the peak of the grey level plot, +6.54% at 10% below the peak, and f5.61% at 20% below. As there is no ‘gold standard’ measurement to compare results, the following criteria was used to select thle most appropriate threshold. On the presumption that there would be minimal inter session variability, vascularity on session one should closely resemble session two. Taking into consideration the 95 % confidence interval for repeatability and the regression lines in Table 2, it was concluded that thresholds measured at 10% below the peak gave the best agreement between visits 1 and 2. Objective assessment of vascularity for contact lens wearers For each contact lens wearing candidate several pictures (2-7) were taken of the dextro-lateral-bulbar conjunctivae Table 2. Mean vascularity values and mean subjective grading for the two contact lens wearing groups. Visit 1 -pre contact lens fitting; visit 2-after 1 month of lens wear; visit 3-after 4 months of lens wear; visit 4-after 10 months of contact lens wear Examination
Visit Visit Visit Visit
1 2 3 4
Visit Visit Visit Visit
1 2 3 4
Percentage of hyperaemia at 70% below + SE SCL group 30.25% + 30.37% k 32.81% 5 30.24% +
(n = IO) 2.28 2.04 2.03 1.64
RGPCL group (n = IO) 28.40% f 2.49 31.11%*1.42 38.77% f 1.28 27.23% f 2.06
Mean subjective grading k SE
1.4kO.16 1.7 f 0.26 1.9 + 0.23 2.1 +0.31 1.3*0.15 1.4kO.22 1.7+0.30 2.5+0.27
8
7
9
10
of vascularity
for (a) peak value and (c) 20% below the means and the
on each examination. Results of the analysis were averaged to produce one result per patient per examination. Mean values were calculated for each visit, for each contact lens wearing group. Table 2 lists the mean values for each lens group, with calculations made at 10% below. Standard errors (SE) are also included. Figure 6 shows the SCL and RGPCL groups’ mean values for each visit with thresholds at 10 % below the peak. Correlation analysis was performed on only four data points. A second order polynomial regression line was fitted. For the SCL group this showed an apparent increase in vascularity after 4 months of contact lens wear, which appeared to have reversed by the 10 month follow up examination. This analysis would suggest a continued reduction in vascularity, after 10 months of contact lens wear. However, such extrapolations cannot be made. It is suspected that vascularity would level out after the final examination. A one way analysis of variance (ANOVA) with the least significant difference multiple comparisons procedure, did not reveal significant difference between visits for vascularity levels. Correlation analysis was also performed on the RGPCL group. A second order polynomial regression line was fitted. This showed an apparent increase in vascularity (hyperaemia) after 4 months of contact lens wear, which appeared to have reversed by the 10 month follow up examination. A one way analysis of variance (ANOVA) with the least significant difference multiple comparisons procedure, revealed a significant difference between visits for vascular@ levels. Values of hyperaemia were significantly greater after 4 months of RGPCL wear, than baseline values, after one month and 10 months of RGPCL wear (P = 0.0001). Although this polynomial increase in vascularity is significant only data from 4 visits are available, hence results must be interpreted cautiously. The polynomial would suggest a continued reduction in
436
Ophthal.
Physiol.
Opt. 1996 16: No 5 RGPCL
.=
25 --
8 z
20 --
c
I :
15.-
E *
lo--
z 70
y = -0.4517~’ + 4.4564x R2 = 0.9881 *
+ 27 879
scLy=0.1072xz+1.109&+29.892 RZ = 0.9081
5-1
07 0
1
2
3 Months
4
5 of contact
6
7
a
9
10
lens wear
Figure 6. SCL (diamond symbols) and RGPCL (square symbols) wearing groups with mean values for each visit. Error bars indicate + standard errors. Equation of the second order polynomial regression line (y = mx2 + nx + c), and R2 values are given for each group. Solid line = RGPCL group, dotted line = SCL group. The one value marked with an asterisk showed statistically significant hyperaemia.
vascularity after 10 months of contact lens wear. However, as with the SCL group such extrapolations cannot be made. Subjective assessmentof vascularity for contact lens wearers The lateral-bulbar conjunctivae was also subjectively graded for hyperaemia. For each patient one grading was taken per examination. Results for each group (SCL and RGPCL) were averaged to give one value per examination, for each lens group (Table 2). When subjectively grading the conjunctivae the examiner was aware of previous readings. An increase in subjective hyperaemia was observed for both lens groups after each examination. However, a Friedman one way ANOVA was performed and in all cases no statistically significant difference in vascularity between visits was found (P = 0.237, P = 0.103 for the SCL and RGPCL group, respectively). The use of a subjective grading system with 5 broad classifications of vascularity may, on average, show hyperaemia but such change is unlikely to be statistically significant. Discussion Quantitative assessment of conjunctival hyperaemia has potential applications in detecting vascular abnormality, and as an index of biocompatibility of contact lenses and ophthalmic medication. For instance, with the introduction of multi-purpose contact lens cleaning solutions accurate monitoring of conjunctival vasculature may give an indication of their long term biocompatibility. The technique described demonstrates acceptable repeatability with 95 % confidence intervals for inter-session measurements of +8.58/-3.95% (Z +1.46mm*/-0.672mm2,
i.e. percentage of 17mm’) when the threshold is taken at 10% below the peak of the grey level histogram. However, if a clinically significant difference lies within this range it will not be detected by this method. Intra-session 95% confidence intervals lie between +6.54% (= * 1.11 mm*). Diurnal and long term fluctuations in conjunctival vasculature may occur. Measurements may be affected by slight variations in the time of examination, or more importantly the number of waking hours before examination. Individuals studied were all members of the UK fire service who work night and day shifts. Some individuals performed a night shift before examination, and may have been awake for 24 hours or more. These are confounding factors when assessing the repeatability of the technique. Selection of an appropriate threshold in the demarcation between vasculature and sclera in the binary image is of importance. As the threshold was decreased from the 10% below peak position to the 20% below, so the episcleral vasculature content in the binary image was seen to reduce. It is suggested that the lower threshold values may give a better representation of the anterior conjunctival plexus. Although both subjective and objective assessment of conjunctival vasculature showed an increase with progressive SCL wear, within this sample, these did not reach statistical significance. Thus, on average, the use of SCL in these patients had a minimal effect on the vasculature. Subjective assessment of vascularity showed an apparent increase for both lens groups but this was not statistically significant. Subjective assessment is susceptible to observer bias, and large inter-observer variations. Computer assisted evaluation of vascularity showed a statistically significant increase in hyperaemia of the lateral-bulbar conjunctivae after 4 months of RGPCL wear. It is proposed that this
Quantifying
phenomenon is due to mechanical adaptation to the lens presence. Firefighters present a group who wear their lenses in unusual ambient environments, and are subject to excessive foreign body involvement (Owen et al., 1995b). Hence the hyperaemic increase may be accentuated by environmental hazards. A hyperaemic response of the dextro-lateral-bulbar conjunctivae is predicted after 4 months of contact lens wear. This will represent an increase in vascular surface area compared to baseline levels by one third (i.e., 36.4% increase in vascular surface area). Hyperaemia of this region of the bulbar conjunctivae will return to normal between 4 to 10 months. A system has been described which provides objective quantification of vascularity of the bulbar conjunctivae. Sensitivity of this technique is considered sufficient to detect clinically significant changes. One disadvantage of the system is the need for photographic images, which does not allow for instant results. As with the current study images can be lost when undertaking photographic reproduction. With the use of a digital camera such difficulties may be overcome. Future work will include studying diurnal and seasonal variations, age related change, and the effects of vasodilator and vasoconstrictor agents on the conjunctival vasculature. Acknowledgements We wish to thank Dr. Alicja Rudnicka for her statistical assistance. Our thanks to Bausch and Lomb for providing all costact lenses fitted, and RGPCL solutions. Also Allergan for supplying the SCL solutions. Lucida version 2 software is available from Kinetic Imaging Ltd, South Harrington Building, Sefton Street, Liverpool, L3 4BQ, UK. References Bland, J. M. and Altman, D. G. (1986). Statistical methods for assessing agreement between two methods of clinical measurement. Lancer i, 307-310. Chaudhuri, S., Chatterjee, S., Katz, N., Nelson, M. and Goldbaum, M. (1989). Detection of blood vessels in retinal images using two-dimensional matched filters. IEEE Trans. Med. Imug. 8, 263-269. Chen, P. C., Kovalcheck, S. W. and Zweifach, B. W. (1987). Ana.ysis of microvascular network in bulbar conjunctiva by imaging processing. ht. J. ofinicrocirc. Clin. Exp. 6, 245-255. Chen, P. C., Li, S. G. and Zweifach, B. W. (1986). Comparisons of morphometric alterations in human bulbar conjunctiva microvascular network in hypertension and Exp. 5, 200. diabetes. Int. J. of Microcirculation-Clin. Coget, J. M., Dupuisuvny, C. H. and Merlen, J. F. (1989). Value of the study of the ocular conjunctivae in screening in diabetes. J. des Maludes Vascularies 14, 68-70. Davydova, N. G., Katsnelson, L. A. and Gurtovaya, Y. Y. (1990). Changes of the bulbar conjunctivae is essential arterial-hypertension. Vestnik Ofthamologi 5, 37-39. Ditzel. J. and Sagild. U. (1954). Morphological and hemodynamic changes in the smaller blood vessels in diabetes mellitus. NeLtxEng. J. Med. 250, 587-594.
vasculature
changes:
C. G. Owen
et al.
437
Dixon, J. M. and Lawaczeck, E. (1963). Cornea1 vascularisation due to contact lenses. Arch. Ophthalmol. 69, 72-75. Kovalchek, S. W., Chen, P. and Zweifach, B. W. (1984). Microvascular changes associated with age in the bulbar conjunctivae. Int. J. of Microcirculation-Clin. Exp. 3, 530. McMonnies, C. W., Chapman-Davies, A. and Holden, B. A. (1982). The vascular response to contact lens wear. Am. J. Optom. Physiol. Opt. 59, 795-799. McMonnies, C. W. and Chapman-Davies, A. (1987). Assessment of conjunctival hyperaemia in contact-lens wearers: Part I. Am. J. Optom. Physiol Opt. 64, 246-250. McMonnies, C. W. and Ho, A. (1991). Conjunctival hyperaemia in non-contact lens wearers. Actu Ophthalmol. 69, 799-801. Miles, F. P. and Nuttall, A. L. (1993). Matched filter estimation of serial blood vessel diameters from video images. ZEE Trans. Med. Imag. 12, 147-152. Owen, C. G., Palmer, J. R., Fitzke, F. W. and Woodward, E. G. (1995a). A new objective method for quantifying hyperaemic changes of the bulbar conjunctivae associated with contact lens wear in adverse environments. Invest. Ophthalmol. Vis. Sci. 36, S313. Owen, C. G., Margrain, T. H. and Woodward, E. G. (1995b). Aetiology and prevalence of eye injuries in the UK fire service. Eye 9(suppl.), 54-59. Palmer, J. R., Owen, C. G., Ford, A. M., Jacobson, R. E. and Woodward, E. G. (1996). Optimal photographic imaging of the bulbar conjunctival vasculature. Ophthal. Physiol. Opt. 16, 144-149. Paton, D. (1961). The conjunctival sign of sickle cell disease. Arch. Ophthalmol. 66, 90-94. Ruben, M., Brown, N., Lobascher, D., Chaston, J. and Morris, J. (1976). Clinical manifestations secondary to soft contact lens wear. Br. J. Ophthalmol. 60, 529-531. Rudnicka, A. R., Steele, C. F., Crabb, D. P. and Edgar, D. F. (1992). Repeatability reproducibility and intersession variability of the Allergen Humphrey ultrasonic biometer. Acta Ophthalmol. 70, 327-334. Sejeant, G. R., Sejeant, B. E. and Condon, P. I. (1972). The conjunctival sign of sickle cell disease. A relationship with irreversibly sickle cells. J. Am. Med. Assoc. 219, 1428-1431. Spencer, T., Phillips, R. P., Sharp, P. F. and Forrester, J. V. (1992). Automated detection and quantification of microaneurysms in fluorescein angiograms. Graefe’s Arch. Clin. Exp. Ophthalmol. 230, 36-41. Tamura, S., Okamoto, Y. and Yanashima, K. (1988). Zerocrossing interval correction in tracing eye-fundus blood vessels. Pattern Recognition 21, 227-233. Villumsen, J. and Alm, A. (1989). Prostaglandin F2a-isopropylester eye drops: effects in normal human eyes. Br. J. Ophthalmol. 73, 419-426. Villumsen, J., Rinquist, J. and Alm, A. (1991). Image analysis of conjunctival hyperaemia. Actu Ophthalmol. 69, 536-539. Willingham, F. F., Coggins, J. M., Cohen, K. L., Goldstein. G. M., Ogle, J. W. and Tripoli, N. K. (1994). Automatic measurement of ocular hyperaemia and external vasculature. Invest. Ophthalmol. Vis. Sci. 35, S1530. Willingham, F. F., Coggins, J. M., Cohen, K. L., Goldstein. G. M., Tripoli, N. K. and Ogle, J. W. (1995a). Hyperaemic measurements and validity in a new digital image analysis system. Invest. Ophthalmol. Vis. Sci. 36, S38. Willingham, F. F.. Cohen, K. L., Coggins, J. M., Tripoli, N. K., Ogle, J. W. and Goldstein, G. M. (1995b). Automatic quantitative measurement of ocular hyperaemia. Curr. Eye Res. 14. 1101-l 108.
Ophthal. Physiol. Opt. Vol. 16, No. 5, pp. 430-437, 1996 0 1996 The College of Optometrists. Published by Elsevier Science Ltd Printed in Great Britain 0275-5408196 $15.00 + 0.00
PII: SO2755408(96)00037-3
A new computer assisted objective method for quantifying vascular changes of the bulbar conjunctivae Christopher
G. Owen,’
’ Applied Goswell London,
Research Centre, City University, Dame Alice Owen London ECIV 7DD, UK; ‘Institute of Ophthalmology, Bath Street, London ECIV 9EL, UK
Vision Road, 11-43
Fred W. Fitzke2
and E. Geoffrey
Woodward’ Building, 31 I-321 University College
Summary A novel computer software method was used to quantify the conjunctival plexus on the scleral background for measurement of the vascular surface area from photographs. A previously described method was used (Palmer, J. R., Owen, C. G., Ford, A. M., Jacobson, R. E. and Woodward, E. G. (1996). Optimal photographic imaging of the bulbar conjunctival vasculature. Ophthal. Physiol. Opt. 16, 144- 149) to optimise photographic imaging of the bulbar conjunctival vasculature by increasing the information content in the image. Repeatability of this technique was evaluated. Twenty subjects (20 eyes) free from ophthalmological and systemic abnormality were examined on two separate occasions. The maximum 95% confidence limits for repeatability are + 8.58/ - 3.95%. For 10 consecutive estimates of vascularity the maximum 95% confidence interval lie between +6.54%. To evaluate the technique the lateral-bulbar conjunctivaein 10 soft (SCL) and 10 rigid gas permeable contact lens (RGPCL) wearers during the first 10 months of contact lens wear, were assessed and compared with subjective grading of hyperaemia. The new method showed sufficient sensitivity in detecting increased hyperaemia in the RGPCL wearing group and demonstrated statistically significant change. Subjective graded assessment of vascularity (using established classifications) detected increased hyperaemia, however, this was not statistically significant. Conjunctival vasculature is a dynamic structure and a source of valuable quantitative information where the ocular environment is varied, or where the ocular surface is affected by disease. Hence it is worthy of further investigation. A simple inexpensive method of computer assisted determination of vascularity is described. Copyright 0 1996 The College of Optometrists. Published by Elsevier Science Ltd.
(1989) found that existing edge detection algorithms for detecting retinal vasculature were unsatisfactory. The grey level profile of a cross section of a blood vessel was approximated by a Gaussian shaped curve. A total of 12 different Gaussian matched filters were devised and used to detect vessel segments. Tamura et al. (1988) used a similar technique to detect retinal vasculature with a spatial filter derived from a second order Gaussian filter. Vasculature detection has also been conducted in the cochlea to determine whether changes in blood flow may be related to hearing loss (Miles and Nuttall, 1993). The conjunctival plexus consists of a multi-layered intricate vascular pattern which is readily observed through its transparent mucus membrane. Its vascular supply to the perilimbal region (4 to 6 mm from the limbus) of the bulbar conjunctivae originates from the ophthalmic artery and
Introduction Examination of vasculature has been considered as a useful source of information for many disorders. Numerous sites have been studied where vasculature can be readily observed. Increasingly, computational methods have been devised to quantify vasculature. Automated approaches have been developed to detect retinal vasculature (Chaudhuri et d., 1989). In 1992 Spencer et al. quantified retinal microaneurysms using digital image processing. Chaudhuri et al. Correspondence and reprints requests ro: Mr. C. G. Owen, Research Fellow, Applied Vision Research Centre, Department of Optometry and Visual Science, Dame Alice Owen Building, 311-321 Goswell Road, London EClV 7DD, UK. Received: 27 March 1996 Revisedfom: 8 May 1996
430
Quantifying
traveis to the limbus via the anterior cilary artery. Systemic disorders such as diabetes (Ditzel and Saglid, 1954; Coget et al., 1989), hypertension (Davydova et al., 1990), and sickle cell disease (Paton, 1961; Sejeant et al., 1972) can affect the morphological and morphometric characteristics of the conjunctival vasculature. Mo~hologica~ alterations in the human bulbar conjunctivae in diabetes are said to include an increase in surface vascularity compared to normals (Chen et al., 19X6), as opposed to hypertensive changes which have diametrically opposite trends, including a decrease in vascularity, length, diameter, large vessels and terminal vessels, compared to normals (Chen et ai., 1986). Such findings may be confounded by vasculature changes associated with age (McMonnies and Ho, 1991; Kovalchek et al., 1984). Ocular inflammation, cornea1 hypoxia, and ocular xerosis may also affect the hyperaemic state of the bulbar conjunctivae. According to McMo~ies and ~apman-Davies (1987) conjunctival hyperaemia occurs ‘commonly’ in contact lens wearers. Chronic conjunctival hyperaemia develops in response to prolonged toxic or physiological insult, or surface deposits or other forms of lens degradation. However, Dixon and Lawaczeck (1963) report that frequent ‘overfilling’ of limbal vessels is common place with asymptomatic hard contact lens wear. Ruben et al. (1976) report increased bulbar conjunctivae hyperaemia with asym,ptomatic soft contact lens wear. McMonnies et al. (1982) conclude that soft contact lens wear induces chronic inferior limbal injection which is not seen with hard contact lens wear. Current techniques to assess hyperaemia of the bulbar conjunctivae, principally rely on subjective assessment, with judgements made from inspection of graded photographic plates (McMonnies and Chapman-Davies, 1987). Such ordinal ranking systems will only allow measurements to be made within certain ‘bracketed’ amounts, thus reducing its sensitivity. Such methods may be open to subjective bias as well as inter subject variabilities such as pupil size, iris colour, palpebral aperture size and skin colour. Investigation of the effect of these physiological variations on the subjective gradings of hyperaemia has never been investigated. A number of computer assisted techniques of conjunctival hyperaemic assessment have been described (Chen et al., 1987; Villumsen et al., 1991; Willingham et al., 1994, 1995a, 1995b). Chen et al. (1987) used a Nikon F3 35 mm camera with macro lens giving a 4.5 x magnification in the film plane. Black and white 35mm film images (ASA 50) containing 6.7 mm2 of conjunctival area were digitised to a 128 x 128 pixel image. This was averaged using a 5 x 5 pixel square and any background luminosity gradients were corrected. The enhanced image was subtracted from the original image, averaged again, and thresholded to isolate vasculature. Villumsen et al. (1991) stored colour transparencies of the temporal bulbar conjunctivae on to a video camera. Video signals were then converted into a mono-
vasculature
changes:
C. G. Owen
et al.
431
chrome 270 x 300 pixel image. A 3 x 3 pixel square was then averaged and a simple edge detection filter applied. The resultant image was then thresholded using a manually selected pixel intensity of a vessel region. This technique was devised to avoid a laborious, time consuming manual point counting system (Vill~sen and Alm, 1989) employed to monitor the conjunctival vascular response to topical application of prostraglandin F,,-isopropylester eye drops. Willingham et al. (1995b) described an automated qualitative measurement of conjunctival hyperaemia using colour images. Images were obtained using a Nikon FS-2 photoslit lamp and recorded to a video image, with a region of interest (ROI) of 3.8 cm*. Using dedicated software the images, stored on magnetic disc (640 pixels x 480 pixels), were automatically masked to exclude specular reflections, lashes, lids and un-illuminated regions. The red (R), green (G), and blue (B) colour content of each pixel was extracted. For each colour the intensity ranged from 0 (darkest) to 255 (brightest). The total intensity (I) of a pixel was the sum of the RGB components. Mean relative redness (RR) was calculated for each pixel from the ratio of the intensity of R to I. The measure of RR was the sum of these ratios divided by the number of active pixels multiplied by 100. To compute vessel area (VA), red free images (G) were selected because of superior vessel contrast. Five Gaussian spatial filters (SD 1, d2, 2, 242, 8 pixels) were applied. Blurring of images allowed correction of luminance gradients over the active pixel region. Vessel and non-vessel areas were rn~u~ly labelled. Typical pixel patterns in feature space spanning a vessel border were modelled by multivariate Gaussian densities. This model was then used to isolate vasculature over the ROT. Once a binary image is obtained morphological and morphometric characteristics can be obtained. For the purposes of vascular examination with contact lens wear it is vessel area which is of most importance, assuming vascular morphology remains unchanged. We describe a simple and inexpensive technique of objectively quantifying vessel area from monochrome images which contain high information content. Unlike previous methodologies of automated vascular examination analysis, repeatabili~ measurements of the current technique were assessed. This technique was used to assess the vascular response of the lateral bulbar conjunctivae in asymptomatic soft contact lens and rigid gas permeable contact lens wear. Palmer et al. (1996) have investigated the optimum colour filter required for imaging the conjunctival vasculature using the Nikon FS-2 photo-slit lamp and Ilford HP5 film. Optimum filtration enables images of superior information content to be obtained. This increases the quality of the binary image generated by the computer programme (Owen et al. f 1995a). If any component of the system were to change, e.g. the use of a charged couple device (CCD) detector instead of HP5 film, the optimum filter would need to be re-calculated.
432
Materials
Ophrhal.
Physiol.
Opt. 1996 16: No 5
and methods
The experimental set up described by Palmer et al. (1996) was used in the current study. Images of the dextro temporal bulbar conjunctivae were captured using a Nikon FS-2 photo-slit lamp with a xenon flash tube, on Ilford HP5 monochrome film. The angle between observation and illumination systems was set to 30”, with the optical axis being approximately normal to the curved conjunctival surface. This has been experimentally found to minimise specular reflections from the conjunctival epithelium (Palmer et al., 1996). Photographic images were scanned using a Polaroid SprintScan 35mm Slide Scanner and saved as a tagged image format file (TIFF). A square region of interest (ROI) immediately lateral to the limbus of 17mm* was manually selected tangential to the limbal conjunctival border. All images were scanned in an identical manner to a 336 x 336 pixel square to a resolution of 133 dots per inch. The 8 bit, 256 grey level image was spatially averaged fifteen times consecutively (using a 3 x 3 pixel square on each occasion), and then subtracted from the original image spatially averaged once. Spatially averaging in a repetitive sequence (X 15) will smooth all high contrast gradients, such as vasculature, leaving low contrast gradients of the background relatively unchanged. When this is subtracted from an image averaged only once, the high contrast gradients (vasculature) remain, Contrast enhancement was performed, which merely allows improved visualisation of the images but does not affect the raw data used for analysis. This was experimentally achieved by multiplying the resultant by 4, and by adding a constant to each pixel grey level using Lucida version 2 software. Figures 1 (a) and (6) show the image of the conjunctival vasculature pre and post processing. Once the image is processed, a frequency distribution of the grey level of each pixel in the image is plotted (see Figure 2). This plot represents the addition of the scleral and conjunctival vasculature grey level profiles. Threshold selection is of importance as this determines the degree of vascular representation. Thresholds were taken at the peak, 10% below the peak grey level, and 20% below the peak (a)
lb)
py’ 0
/ 50
100
Grey level (O-256) Figure 2. Hypothetical frequency distribution plot of each pixel grey level in the image. Thresholds for the binary image are taken at the peak, 10% below, and 20% below as indicated with vertical solid lines right to left respectively. The frequency distribution represents the combined scleral and vasculature grey level profiles. Thick solid line = resultant grey level profile, dashed line = hypothetical scleral content, and dotted line = hypothetical vasculature content.
grey level, as the cut-off between vasculature and sclera to create the binary image (see Figure 3). Both the episcleral and conjunctival plexus will be represented in differing amounts. Selection of the most appropriate threshold is given in the results section. Once a binary image is achieved the percentage of vasculature within the 17 mm2 area is calculated. This is equivalent to a dimensional value in terms of the percentage area of the ROI (17 mm’) which is considered as vasculature. Repeatability Twenty subjects (20 right eyes) free from ophthalmological and systemic abnormalities were photographed on 2 separate occasions to evaluate repeatability. Age ranged from 30 to 53 years (mean 41.5 years). All subjects were non contact lens wearers and were seen at the same time of day for each examination (k 1% hours) approximately 3 weeks apart. On arrival a questionnaire was completed, symptoms and history recorded. This allowed sufficient time to adapt to the ambient environment. This was followed by slit-lamp photography of the dextro-lateral bulbar conjunctiva. A full anterior eye examination was carried out including subjective grading of hyperaemia and fluorescein staining. Statistical analysis
Figure 1. (a) and (b): Images of the lateral-bulbar junctivae pre and post image processing, respectively.
con
Proportion of vasculature in the image (expressed as a percentage) was calculated at (a) the peak of the grey level
Quantifying
vasculature
changes:
C. G. Owen
et al.
433
(b)
(a)
Figure 3. Shows a binary image obtained (b) 10%
below
the peak
grey
level,
for threshold taken at (a) the peak of the grey level histogram, and (c) 20% below the peak grey level.
plot, ;b) 10% below, and (c) 20% below. These were analysed separately comparing sessions 1 and 2. (Graphs were drawn for pairs of data measured at each level (a, b and c), Correlation coefficients were evaluated as a measure of the association between pairs of data. Proximity of the points to the regression line gives some indication of the agreement. An association where the regression line intercepts with the origin, and has a gradient of one, was thought to offer the most suitable threshold. A preferred alternative to the correlation approach is to consider the difference between paired values obtained at session 1 and 2, plotted against their mean for each vascularity value (Rudnicka et al., 1992). These plots allow investigation of any relationship between the measurement error and true value. When the true value is unknown a mean of the measurements is the best estimate (Bland and Altman, 1986). The coefficient of repeatability can be calculated, defined by the British Standards Institute (BS5497) to be twice the standard deviation (SD) from the mean difference. By definition, 95 % of the values will lie between +2 SD. Provided that the differences within *2 SD would not be clinically important one can say there is good repeatability between the two measures. Ideally the threshold selected should have the smallest confidence range. Conjunctival vasculature is a dynamic structure, hence 10 consecutive photographs of the same eye were taken. The rationale being that there would be minimal vasculature change over the short period of photography. The mean f2 SD was calculated to give another measure of variance in the estimated percentage of vasculature. This gives an indication of intra-session variabilities. Objective vs subjective assessment of hyperaemia Vascularity of the dextro-lateral-bulbar conjunctiva in 10 soft (Bausch and Lomb 70) and 10 rigid (Bausch and Lomb Quantum) daily contact lens wearers was evaluated by this novel technique, and compared with subjective grading. Measurements were taken pre contact lens fitting, after
1 month, 4 months, and 10 months of contact lens wear. Patients were all members of the UK fire service. All aspects of both the bulbar and palpebral conjunctivae were examined and graded. Vascular&y of the bulbar conjunctivae was subjectively classified from 1 to 5 where 1 represents normal vascularity, and 5, gross hyperaemia and extreme vessel dilation. This classification was based on the method described by McMonnies and Chapman-Davies (1987) with reference to photographic plates. Although these showed the inferior aspect of the bulbar conjunctivae, the vascular appearance was extrapolated to the lateral portion. As changes in normal contact lens wear are likely to be small, a grading of 1 to 5 was thought to be sufficient. In addition to the subjective measurement, observed over the entirety of the lateral conjunctivae, a masked area (17 mm’) immediately adjacent to the limbus was objectively quantified, by the method described above. Direct comparison of the 2 methods, although useful, should be treated cautiously as differing areas of vascularity were assessed. Results Repeatability Data obtained at session 2 was plotted against session 1, for 15 eyes, measured at (a) the peak value, (b) 10% below, and (c) 20% below the peak. Five eyes were rejected due to an experimental oversight. Correlation analysis followed by linear regression by the least squares method was performed. Values for R2 and equation of the regression lines are given in Table 1. Figure 4 shows the difference between paired values obtained at session 1 and 2, plotted against their mean hyperaemic value as measured at each threshold. This allows a more valid assessment of agreement. Individual data points show an acceptable spread about the mean at all x values, with no apparent ‘funnelling effect’. The 95 % confidence limits for inter-session repeatability are +9.85/ -8.51% (spread 18.36%) when measured at the peak of
434
Ophthal.
Physiol.
Opt.
1996
16: No 5
the grey level distribution plot, +8.58/-3.95% (spread 12.53%) at 10% below the peak, and +6.26/-5.98% (spread 12.24%) at 20% below. Figure 5 shows 10 consecutive estimates of vascularity, as measured at each threshold level. The mean value is indicated in each case by the dashed line, and the solid lines represent f 2 SD from the mean. The 95 % confidence
Table 1. Values for R2 and equation of the regression line measured at the peak threshold, 10% below, and 20% below Measuring
level
Equation
of regression
y = 0.7976x y = 1.0248x y = 0.6766x
Peak threshold 10% Below peak 20% Below peak
R2 value
line
0.4221 0.757 0.5393
+ 12.95 + 1.6664 + 3.9854
(a)
50
lb)
60
55
Mean vascularity
percentage
65
70
75
as measured at the peak histogram level (%),
10 A
T@
8--
b s
6--
25
4--
% E 3 2 1
2
-2 --
.c
-4
8 ii
-6 --
z g
d
* d
A d
A d
A
o--
A
A
a
A A
A
-8 -I
-10 7 20
15
10
Mean vascularity
percentage
35
30
25
40
as measured at 10% below the peak (%).
(c’ &‘j 0 0
t .:
2 --
&
o-
0 0
0 00
a !i .E ' Es $ g
0
-2 -4
-6-
0
0
0
-8 -10 7 0
I 2
4
Mean vascularity
6
8
percentage
10
12
14
16
18
20
as measured at 20% below the peak (%).
Figure 4. Differences between paired values of vascularity for sessions 1 and 2 plotted against their mean (n = 15). Thick solid line represents mean difference and the thin solid line 95% confidence limits, where (a) is measured at the peak, (b) 10% below the peak, and (c) 20% below the peak.
Quantifying
; .z
70
‘g;
60 50
0
vasculature
0
0
5
6
changes:
C. G. Owen
et al.
435
0
D
0 ‘ij I k 8 * z 0 -e
40 30 20 10 0 0
1
3
2 Number
4 of consecutive
estimates
Figure 5. Ten consecutive estimates of vascularity (open circles); (b) 10% below peak (open triangles), oeak (ooen sauares). The dashed line represents solid line +-2SD.
limits are * 10.93 % when measured at the peak of the grey level plot, +6.54% at 10% below the peak, and f5.61% at 20% below. As there is no ‘gold standard’ measurement to compare results, the following criteria was used to select thle most appropriate threshold. On the presumption that there would be minimal inter session variability, vascularity on session one should closely resemble session two. Taking into consideration the 95 % confidence interval for repeatability and the regression lines in Table 2, it was concluded that thresholds measured at 10% below the peak gave the best agreement between visits 1 and 2. Objective assessment of vascularity for contact lens wearers For each contact lens wearing candidate several pictures (2-7) were taken of the dextro-lateral-bulbar conjunctivae Table 2. Mean vascularity values and mean subjective grading for the two contact lens wearing groups. Visit 1 -pre contact lens fitting; visit 2-after 1 month of lens wear; visit 3-after 4 months of lens wear; visit 4-after 10 months of contact lens wear Examination
Visit Visit Visit Visit
1 2 3 4
Visit Visit Visit Visit
1 2 3 4
Percentage of hyperaemia at 70% below + SE SCL group 30.25% + 30.37% k 32.81% 5 30.24% +
(n = IO) 2.28 2.04 2.03 1.64
RGPCL group (n = IO) 28.40% f 2.49 31.11%*1.42 38.77% f 1.28 27.23% f 2.06
Mean subjective grading k SE
1.4kO.16 1.7 f 0.26 1.9 + 0.23 2.1 +0.31 1.3*0.15 1.4kO.22 1.7+0.30 2.5+0.27
8
7
9
10
of vascularity
for (a) peak value and (c) 20% below the means and the
on each examination. Results of the analysis were averaged to produce one result per patient per examination. Mean values were calculated for each visit, for each contact lens wearing group. Table 2 lists the mean values for each lens group, with calculations made at 10% below. Standard errors (SE) are also included. Figure 6 shows the SCL and RGPCL groups’ mean values for each visit with thresholds at 10 % below the peak. Correlation analysis was performed on only four data points. A second order polynomial regression line was fitted. For the SCL group this showed an apparent increase in vascularity after 4 months of contact lens wear, which appeared to have reversed by the 10 month follow up examination. This analysis would suggest a continued reduction in vascularity, after 10 months of contact lens wear. However, such extrapolations cannot be made. It is suspected that vascularity would level out after the final examination. A one way analysis of variance (ANOVA) with the least significant difference multiple comparisons procedure, did not reveal significant difference between visits for vascularity levels. Correlation analysis was also performed on the RGPCL group. A second order polynomial regression line was fitted. This showed an apparent increase in vascularity (hyperaemia) after 4 months of contact lens wear, which appeared to have reversed by the 10 month follow up examination. A one way analysis of variance (ANOVA) with the least significant difference multiple comparisons procedure, revealed a significant difference between visits for vascular@ levels. Values of hyperaemia were significantly greater after 4 months of RGPCL wear, than baseline values, after one month and 10 months of RGPCL wear (P = 0.0001). Although this polynomial increase in vascularity is significant only data from 4 visits are available, hence results must be interpreted cautiously. The polynomial would suggest a continued reduction in
436
Ophthal.
Physiol.
Opt. 1996 16: No 5 RGPCL
.=
25 --
8 z
20 --
c
I :
15.-
E *
lo--
z 70
y = -0.4517~’ + 4.4564x R2 = 0.9881 *
+ 27 879
scLy=0.1072xz+1.109&+29.892 RZ = 0.9081
5-1
07 0
1
2
3 Months
4
5 of contact
6
7
a
9
10
lens wear
Figure 6. SCL (diamond symbols) and RGPCL (square symbols) wearing groups with mean values for each visit. Error bars indicate + standard errors. Equation of the second order polynomial regression line (y = mx2 + nx + c), and R2 values are given for each group. Solid line = RGPCL group, dotted line = SCL group. The one value marked with an asterisk showed statistically significant hyperaemia.
vascularity after 10 months of contact lens wear. However, as with the SCL group such extrapolations cannot be made. Subjective assessmentof vascularity for contact lens wearers The lateral-bulbar conjunctivae was also subjectively graded for hyperaemia. For each patient one grading was taken per examination. Results for each group (SCL and RGPCL) were averaged to give one value per examination, for each lens group (Table 2). When subjectively grading the conjunctivae the examiner was aware of previous readings. An increase in subjective hyperaemia was observed for both lens groups after each examination. However, a Friedman one way ANOVA was performed and in all cases no statistically significant difference in vascularity between visits was found (P = 0.237, P = 0.103 for the SCL and RGPCL group, respectively). The use of a subjective grading system with 5 broad classifications of vascularity may, on average, show hyperaemia but such change is unlikely to be statistically significant. Discussion Quantitative assessment of conjunctival hyperaemia has potential applications in detecting vascular abnormality, and as an index of biocompatibility of contact lenses and ophthalmic medication. For instance, with the introduction of multi-purpose contact lens cleaning solutions accurate monitoring of conjunctival vasculature may give an indication of their long term biocompatibility. The technique described demonstrates acceptable repeatability with 95 % confidence intervals for inter-session measurements of +8.58/-3.95% (Z +1.46mm*/-0.672mm2,
i.e. percentage of 17mm’) when the threshold is taken at 10% below the peak of the grey level histogram. However, if a clinically significant difference lies within this range it will not be detected by this method. Intra-session 95% confidence intervals lie between +6.54% (= * 1.11 mm*). Diurnal and long term fluctuations in conjunctival vasculature may occur. Measurements may be affected by slight variations in the time of examination, or more importantly the number of waking hours before examination. Individuals studied were all members of the UK fire service who work night and day shifts. Some individuals performed a night shift before examination, and may have been awake for 24 hours or more. These are confounding factors when assessing the repeatability of the technique. Selection of an appropriate threshold in the demarcation between vasculature and sclera in the binary image is of importance. As the threshold was decreased from the 10% below peak position to the 20% below, so the episcleral vasculature content in the binary image was seen to reduce. It is suggested that the lower threshold values may give a better representation of the anterior conjunctival plexus. Although both subjective and objective assessment of conjunctival vasculature showed an increase with progressive SCL wear, within this sample, these did not reach statistical significance. Thus, on average, the use of SCL in these patients had a minimal effect on the vasculature. Subjective assessment of vascularity showed an apparent increase for both lens groups but this was not statistically significant. Subjective assessment is susceptible to observer bias, and large inter-observer variations. Computer assisted evaluation of vascularity showed a statistically significant increase in hyperaemia of the lateral-bulbar conjunctivae after 4 months of RGPCL wear. It is proposed that this
Quantifying
phenomenon is due to mechanical adaptation to the lens presence. Firefighters present a group who wear their lenses in unusual ambient environments, and are subject to excessive foreign body involvement (Owen et al., 1995b). Hence the hyperaemic increase may be accentuated by environmental hazards. A hyperaemic response of the dextro-lateral-bulbar conjunctivae is predicted after 4 months of contact lens wear. This will represent an increase in vascular surface area compared to baseline levels by one third (i.e., 36.4% increase in vascular surface area). Hyperaemia of this region of the bulbar conjunctivae will return to normal between 4 to 10 months. A system has been described which provides objective quantification of vascularity of the bulbar conjunctivae. Sensitivity of this technique is considered sufficient to detect clinically significant changes. One disadvantage of the system is the need for photographic images, which does not allow for instant results. As with the current study images can be lost when undertaking photographic reproduction. With the use of a digital camera such difficulties may be overcome. Future work will include studying diurnal and seasonal variations, age related change, and the effects of vasodilator and vasoconstrictor agents on the conjunctival vasculature. Acknowledgements We wish to thank Dr. Alicja Rudnicka for her statistical assistance. Our thanks to Bausch and Lomb for providing all costact lenses fitted, and RGPCL solutions. Also Allergan for supplying the SCL solutions. Lucida version 2 software is available from Kinetic Imaging Ltd, South Harrington Building, Sefton Street, Liverpool, L3 4BQ, UK. References Bland, J. M. and Altman, D. G. (1986). Statistical methods for assessing agreement between two methods of clinical measurement. Lancer i, 307-310. Chaudhuri, S., Chatterjee, S., Katz, N., Nelson, M. and Goldbaum, M. (1989). Detection of blood vessels in retinal images using two-dimensional matched filters. IEEE Trans. Med. Imug. 8, 263-269. Chen, P. C., Kovalcheck, S. W. and Zweifach, B. W. (1987). Ana.ysis of microvascular network in bulbar conjunctiva by imaging processing. ht. J. ofinicrocirc. Clin. Exp. 6, 245-255. Chen, P. C., Li, S. G. and Zweifach, B. W. (1986). Comparisons of morphometric alterations in human bulbar conjunctiva microvascular network in hypertension and Exp. 5, 200. diabetes. Int. J. of Microcirculation-Clin. Coget, J. M., Dupuisuvny, C. H. and Merlen, J. F. (1989). Value of the study of the ocular conjunctivae in screening in diabetes. J. des Maludes Vascularies 14, 68-70. Davydova, N. G., Katsnelson, L. A. and Gurtovaya, Y. Y. (1990). Changes of the bulbar conjunctivae is essential arterial-hypertension. Vestnik Ofthamologi 5, 37-39. Ditzel. J. and Sagild. U. (1954). Morphological and hemodynamic changes in the smaller blood vessels in diabetes mellitus. NeLtxEng. J. Med. 250, 587-594.
vasculature
changes:
C. G. Owen
et al.
437
Dixon, J. M. and Lawaczeck, E. (1963). Cornea1 vascularisation due to contact lenses. Arch. Ophthalmol. 69, 72-75. Kovalchek, S. W., Chen, P. and Zweifach, B. W. (1984). Microvascular changes associated with age in the bulbar conjunctivae. Int. J. of Microcirculation-Clin. Exp. 3, 530. McMonnies, C. W., Chapman-Davies, A. and Holden, B. A. (1982). The vascular response to contact lens wear. Am. J. Optom. Physiol. Opt. 59, 795-799. McMonnies, C. W. and Chapman-Davies, A. (1987). Assessment of conjunctival hyperaemia in contact-lens wearers: Part I. Am. J. Optom. Physiol Opt. 64, 246-250. McMonnies, C. W. and Ho, A. (1991). Conjunctival hyperaemia in non-contact lens wearers. Actu Ophthalmol. 69, 799-801. Miles, F. P. and Nuttall, A. L. (1993). Matched filter estimation of serial blood vessel diameters from video images. ZEE Trans. Med. Imag. 12, 147-152. Owen, C. G., Palmer, J. R., Fitzke, F. W. and Woodward, E. G. (1995a). A new objective method for quantifying hyperaemic changes of the bulbar conjunctivae associated with contact lens wear in adverse environments. Invest. Ophthalmol. Vis. Sci. 36, S313. Owen, C. G., Margrain, T. H. and Woodward, E. G. (1995b). Aetiology and prevalence of eye injuries in the UK fire service. Eye 9(suppl.), 54-59. Palmer, J. R., Owen, C. G., Ford, A. M., Jacobson, R. E. and Woodward, E. G. (1996). Optimal photographic imaging of the bulbar conjunctival vasculature. Ophthal. Physiol. Opt. 16, 144-149. Paton, D. (1961). The conjunctival sign of sickle cell disease. Arch. Ophthalmol. 66, 90-94. Ruben, M., Brown, N., Lobascher, D., Chaston, J. and Morris, J. (1976). Clinical manifestations secondary to soft contact lens wear. Br. J. Ophthalmol. 60, 529-531. Rudnicka, A. R., Steele, C. F., Crabb, D. P. and Edgar, D. F. (1992). Repeatability reproducibility and intersession variability of the Allergen Humphrey ultrasonic biometer. Acta Ophthalmol. 70, 327-334. Sejeant, G. R., Sejeant, B. E. and Condon, P. I. (1972). The conjunctival sign of sickle cell disease. A relationship with irreversibly sickle cells. J. Am. Med. Assoc. 219, 1428-1431. Spencer, T., Phillips, R. P., Sharp, P. F. and Forrester, J. V. (1992). Automated detection and quantification of microaneurysms in fluorescein angiograms. Graefe’s Arch. Clin. Exp. Ophthalmol. 230, 36-41. Tamura, S., Okamoto, Y. and Yanashima, K. (1988). Zerocrossing interval correction in tracing eye-fundus blood vessels. Pattern Recognition 21, 227-233. Villumsen, J. and Alm, A. (1989). Prostaglandin F2a-isopropylester eye drops: effects in normal human eyes. Br. J. Ophthalmol. 73, 419-426. Villumsen, J., Rinquist, J. and Alm, A. (1991). Image analysis of conjunctival hyperaemia. Actu Ophthalmol. 69, 536-539. Willingham, F. F., Coggins, J. M., Cohen, K. L., Goldstein. G. M., Ogle, J. W. and Tripoli, N. K. (1994). Automatic measurement of ocular hyperaemia and external vasculature. Invest. Ophthalmol. Vis. Sci. 35, S1530. Willingham, F. F., Coggins, J. M., Cohen, K. L., Goldstein. G. M., Tripoli, N. K. and Ogle, J. W. (1995a). Hyperaemic measurements and validity in a new digital image analysis system. Invest. Ophthalmol. Vis. Sci. 36, S38. Willingham, F. F.. Cohen, K. L., Coggins, J. M., Tripoli, N. K., Ogle, J. W. and Goldstein, G. M. (1995b). Automatic quantitative measurement of ocular hyperaemia. Curr. Eye Res. 14. 1101-l 108.