Download Possible errors in the measurement of retinal lesions.

Possible Errors in the Measurement of Retinal Lesions
/ V. Arnold*]. W. C. Gatesj and K. M. Taylor*
Purpose. To determine the accuracy of the measurement of structure in the retina from fundus
photographs.
Method. A model eye was developed to measure the retinal image size of the Zeiss fundus
camera. A X10 microscope objective with a focal length of 16.41 mm was used together with a
microscope graticule.
Results. Under emmetropic conditions the difference between the calculated and the measured magnification was 8.6%. The total change in magnification from myopia to hyperopia
was —24.63% to +18.1%. When the camera position was altered with respect to the model eye
by ±5 mm under myopic conditions the change in magnification was —4.5% and +6.8%. In the
hyperopic condition the change was —5.6% to +4.6%.
Conclusions. Measuring the size of structure in the retina from photographs may therefore
prove to be inaccurate but in some circumstances determining the percentage change in size
may prove more reliable. Invest Ophthalmol Vis Sci. 1993;34:2576-2580.
iJince the introduction of the fundus camera, many
attempts have been made to determine the size of
various structures in the retina. Photogrammetric
methods have also been employed with the fundus
camera to assess the depth of the optic cup. To determine these parameters it is essential to know the exact
magnification of the camera image. Of particular interest to our cardiac surgery unit is to determine the level
of capillary dropout in fluorescein angiograms obtained during cardiopulmonary bypass surgery.1"3 In
these studies, using a method of intraoperative fluorescein angiography, and the subsequent use of digital
image analysis, both developed by our group, 47 we
have shown that capillary dropout is present in the
retina in a large proportion of patients undergoing
cardiopulmonary bypass surgery. Because it is now
possible to not only determine those vessels that reFrom *Cardiac Surgery Unit, Royal Postgraduate Medical School and the
•fDepartment of Surveying and Photogrammetry, University College, London,
United Kingdom.
Supported in part by the British Heart Foundation.
Submitted for publication: August 14, 1992; accepted January 26, 1993.
Proprietary interest catagory: N.
Reprint requests:]. V. Arnold, Cardiac Surgery Unit, Royal Postgraduate Medical
School, Du Cane Road, London, WJ2 ONN, England.
2576
main unfilled throughout the entire sequence automatically, but also to ascertain the time for the delayed
filling of some of the vessels,8 it would seem that a
method of determining the exact length of the vessels,
enabling a direct comparison between large numbers
of patients is desirable. The same method would also
allow the determination of other parameters such as
vessel caliber and areas of lesions. Many attempts
have been made to determine the exact magnification
of the retinal image including those with ametropic
Q-l 9
errors.
The magnification of the the image from the Zeiss
(Welwyn, Garden City) fundus camera is X2.43 for
emmetropic eyes and varies from XI .79 with a refractive error of —16 diopters to X3.1 at +16 diopters
(personal communication, Carl Zeiss, Oberkochen,
July 1990). These figures agree quite closely with those
reported by Bengtsson and Krakau.10 The same figures are obtained if the formula specified by Littman is
used.13 Littman states that the magnification is equal
to the focal length of the fundus camera divided by the
focal length of the eye. This formula can only be expected to apply to emmetropic eyes of varying length
and since most refractive errors are axial, the specified
Investigative Ophthalmology & Visual Science, July 1993, Vol. 34, No. 8
Copyright © Association for Research in Vision and Ophthalmology
Downloaded From: http://iovs.arvojournals.org/pdfaccess.ashx?url=/data/journals/iovs/933174/ on 06/16/2017
2577
Retinal Measurement Errors
magnification with different refractive errors must be
considered suspect. To verify this, we devised a
"model eye" based on a X10 microscope objective,
having a focal length close to that of Gullstrands
model eye and used this to determine changes in magnification that occur under different "ametropic"
conditions and changes in camera position with respect to the "model eye." In a study (unpublished
data) determining the effect of different pharmacologic interventions during cardiopulmonary bypass,
we subjected 26 dogs to retinal fluorescein angiography. At the end of the experiment, the dogs were killed
for histologic studies and the study eye was enucleated
for the assessment of the magnification of the photographic image.
MATERIALS AND METHODS
The animals were treated according to the ARVO recommendations on the use of animals in research.
We used the "magnification" method to measure
the focal lengths of both the fundus camera and the
microscope objective. A scale is placed in front of the
lens to be tested and a sharp image formed, and the
distance between some point on the lens and the image
is measured (V) together with the image size (M). The
object is then moved by a distance and a sharp image
formed and the same parameters are measured again.
The focal length of the lens is then the difference between the two image distances divided by the difference
in image magnification sizes (dV/dM).
To measure the change in magnification with refractive error and variation in camera position a flat
scale divided into 0.1-mm intervals was mounted on a
microscope stage and the microscope objective was
clamped in position directly above it. The fundus camera was then positioned vertically relative to the objective. Water, having a similar refractive index to the
media in the eye (1.336) was introduced between the
objective and the scale. The fundus camera draw tube
was then set to the emmetropic position (infinity). This
was determined with a collimator, which provides an
image of a target at infinity. The scale was transilluminated with an electronic flash synchronized with the
camera. The microscope stage was then adjusted to
provide a sharp image of the scale in the center of the
field. Ideally, this scale should be curved because the
optics of the fundus camera are designed for use with
a curved field so that a flat scale only approximates to
the retina in practice, but since only the central 0.5
mm of the scale was used for measurement, the error
introduced is considered to be very small. Photographs were obtained of the scale at different positions of the draw tube from minimum to maximum
extension and the change in distance of the scale relative to the objective was recorded. The draw tube of
the fundus camera was then set to minimum extension
(myopic condition) and photographs were obtained
when the camera position was altered by ± 5 mm. It
was necessary to change the position of the scale
slightly to achieve sharp focus. This was repeated with
the camera tube at maximum extension (hyperopic
condition). The photographs were recorded on Kodak
Technical Pan film developed in Kodak Dl9 developer
(Eastman Kodak, Rochester, NY) for 4 min at 20°C.
After processing, the negatives were placed in a Zeiss
(Jena) microfilm reader and projected at a magnification of XI 4.8 on the screen. The central 0.5 mm of the
scale was used to determine the magnification at the
image plane of the fundus camera.
To calculate the image magnification of the dog
retina the optic disk was photographed with the fundus camera using red-free illumination. Two photographs were taken, one with a contact lens in place and
one without in 11 of the 26 dogs. The contact lens was
used throughout the study with saline to maintain clarity. The position of the draw tube on the camera was
measured and it was found that the position altered
when the contact lens was removed. This meant that
the contact lens and saline introduced a refractive
error. According to Littman,13 this should not produce a change in image size because the error is refractive rather than axial.
After enucleation, the external axial length was
measured and the eye was dissected and the disk photographed using a Zeiss dissecting microscope with
camera attachment. The magnification at the film
plane with this instrument was X9.1. Because the disks
are of an irregular shape, the area rather than the
width was measured and compared with the area of the
image obtained with the fundus camera. To determine
these areas we used a Contextvision image analysis system (Systems AB Linkopung, Sweden) to trace the
areas on the computer monitor using a mouse-controlled cursor.
RESULTS
The focal length of the microscope objective was
found to be 16.41 mm, which differs from that of
Gullstrands eye by 3.9%. The focal length of the fundus camera was 41.71 mm. According to Littman's
formula, this should provide a magnification of X2.54
with the model eye and X2.44. with gullstrands eye. In
practice, we found that the actual magnification with
the model eye was X2.76, a difference of 8.6% from
that calculated. The change in magnification for the
different ametropic conditions is shown in Figure 1.
The minimum magnification was X2.08 and the maximum X3.26, the change in magnification therefore is
+ 18.1% for hyperopia and —24.63% for myopia. The
change in the position of the scale was 12.02 mm over
Downloaded From: http://iovs.arvojournals.org/pdfaccess.ashx?url=/data/journals/iovs/933174/ on 06/16/2017
2578
Investigative Ophthalmology & Visual Science, July 1993, Vol. 34, No. 8
Calculated verses measured magnification
using model eye
Magnification with and without contact lens
3-
1
2-
Calculated
20
40
60
80
Camera extension (mm)
FIGURE l. The difference between the calculated and the
measured magnification.
3.4
3.6
3.8
4.0
4.2
4.4
4.6
With contact lens
the complete range of the camera draw tube extension, representing an increase of 8.33 mm from the
emmetropic to myopic position and a decrease of 3.69
mm for the hyperopic position.
When the camera was set to the minimum extension of the draw tube, (myopia) and the whole camera
moved ±5 mm with respect to the model eye, the
change in magnification was; —5 mm, XI.99, and +5
mm, X2.23
When the camera was focused for hyperopia, that
is, maximum extension, the change in magnification
was from X3.08 when too far away and X3.41 when
too close (Figure 2). In practice, this degree of error
should not arise if the operator is careful when positioning the camera but it serves to demonstrate the
possible error caused by incorrect camera positioning
when large refractive errors are present.
In the canine eye, The mean magnification was
3.89 with the contact lens and 4.08 without, a mean
Change in magnification with variation
of distance between camera and model eye
Myopic condition
Hyperopic condition
-2
0
2
Camera position
FIGURE 2. The change in magnification in the presence of a
high refractive error when the camera position is adjusted
by ± 5 mm
FIGURE 3. The change in magnification with and without a
contact lens.
difference of 4.8% (Figure 3) This is in contradiction
to Littman13 who states that the image size does not
change when the error is caused by the refractive component.
DISCUSSION
Pach et al14 have previously investigated the changes in
magnification with refractive error, camera decentration, and by varying the distance between the camera
and the eye. Their results are somewhat confusing.
They determined that there was no change in magnification when the camera distance was altered by —1.5
cm and +15 cm. This is hardly surprising because their
subject was emmetropic and therefore the fundus is at
infinity. They also found a change in magnification by
decentering the image when the Zeiss fundus camera
was used. They report an increase in image size of 20%
when the disc was at the edge of the field, in two emmetropic subjects. We randomly selected color photographs of ten eyes taken with the Zeiss fundus camera.
The disk area is sometimes difficult to define, so the
same structure on the disk was measured, at the center
of the field, and at the edge. The refractive condition
of the subjects were unknown. We found a mean increase in image size of 5.5% (range 1.4-9.4%; Fig. 4).
This represents a high degree of error but not as high
as that in Pach's report. Our findings in the case of
refractive ametropia also differ from their results.
We have shown that several factors influence the
determination of the magnification, and therefore the
size of various parameters within the fundus photograph. Several methods have been proposed with
Downloaded From: http://iovs.arvojournals.org/pdfaccess.ashx?url=/data/journals/iovs/933174/ on 06/16/2017
2579
Retinal Measurement Errors
Percentage change in magnification
at edge of field
% change
10-,
86-
++
4-
+
2-
FIGURE 4. The change in magnification relative to position
within the field.
which the magnification may be calculated. The most
common approach seems to be to obtain values of the
axial length, corneal curvature, lens thickness, and the
depth of the anterior chamber. However, Snead et al15
have reported errors in ultrasound measurements for
axial length when hard and soft probes are used. Even
if the calculations were accurate, there could still be
errors introduced by incorrect camera position when
high refractive errors are present. Bengtsson et al10
calculated the change in magnification with refractive
error, based on Gullstrand's model eye. We have
shown that the method is not accurate, partly because
when the calculation is based on the first focal length
of the eye, the error is in fact a change in the axial
length but the eye still remains emmetropic, whereas
most refractive errors are axial. Our results show that
the actual magnification is higher than that calculated
for emmetropic eyes. Baumbach et al9 have reported a
method of "Absolute ocular fundus dimensions" using an interference pattern generated by a laser coupled to the fundus camera. The fringes appear to be
derived from coherent illumination of a regular series
of apertures, however, a transverse equally spaced series of coherent point sources (eg, derived from a
laser) would produce interference in the form of a
family of confocal hyperboloids of revolution. The
pattern produced on a flat screen would have only the
central fringe straight. Outer fringes would be increasingly curved outward, with increasing spacing. The series of secondary sources is produced by multiple reflections in a thin transparent plate coated with highreflection films. The plate appears to be canted at
about 20 degrees to produce the stated transverse separation of the secondary sources, and must therefore
been about 0.25 mm thick. The laser was focused
down on this component, and the original set of secondary sources would have appeared to be formed
behind the primary focused "source," as a conse-
quence of the successive path increments of each double transit of the space between the reflecting surfaces. Consequently, the secondary sources are not
only transversely separated by 31 /zm, but are separated axially also, by 4/3 times the thickness of the
plate, or approximately 300 /im. The effect of such an
oblique but mainly axial separation of sources would
be to produce concentric interference rings of varying
spacing. Because the plate is tilted at such a large angle, the pattern observed is only a small part of the
outer field and appears as nearly equidistant curved
fringes. This seems to be the case in the photographs
shown.
Furthermore, the observed pattern through an aperture close to the secondary source means that the
form of the pattern observed will appear to vary very
little with the distance and form of the surface reflecting the pattern, and is therefore of no use for locating
or studying the height or form of local features in the
surface.
Most importantly, the magnification between the
tilted plate and the retina is not considered: the many
refractions and changes of media make this most complicated.
Furthermore, the calculations used in this study
are based on measurements made on, or very close to
the optical axis. It cannot be assumed that the magnification at the edge of the field is the same at the periphery.
CONCLUSIONS
We have developed a model eye to determine the
changes in image size under different degrees of ametropia and also when the camera position is altered.
The results show that the measured magnification
differs significantly from the calculated data when previously described formulas are used. Our results are
based on measurements made close to the optical axis.
Under varying degrees of ametropia, not only the
length of the eye will change but also the amount of
curvature and in these circumstances the area photographed may no longer form part of a perfect sphere;
therefore, the image size, either measured or calculated on the optical axis, will not apply to the parts of
the image at the edge of the field. This is will add
significantly to the error.
We conclude that most current methods for determining the size of structure in the fundus photograph suffer from significant errors. This will be particularly relevant when attempting to measure blood velocity using the laser Doppler technique in which it is
important to calculate the cross-sectional area of a vessel to determine the volume flow.
One possible solution to this problem is not to
depend on the assumption that the magnification is
Downloaded From: http://iovs.arvojournals.org/pdfaccess.ashx?url=/data/journals/iovs/933174/ on 06/16/2017
2580
Investigative Ophthalmology & Visual Science, July 1993, Vol. 34, No. 8
uniform but to compare differential changes of the
structures in the same part of successive fields. The use
of digital image analysis can help to overcome some of
these problems. We have previously reported a
method of digitizing fluorescein angiograms in which
the detail in two frames obtained on different occasions can be spatially and geometrically registered.4'5
This process automatically corrects for magnification
errors and therefore the two fields selected for analysis will be identical in size regardless of the actual magnification. On this basis, percentage differences can be
calculated and therefore a comparison between patients
is possible. Work is currently in progress by our group
to automatically identify vessels of different sizes to
enable differentiation between larger vessels and capillaries. If digital image analysis is applied to retinal
photographs and automatic spatial and geometric
alignment is achieved, the calculation of the magnification may be alleviated in many cases. This will probably not apply in cases of stereophotogrammetry for
which accurate depth measurements are required. Further research is required before reliable measurements can be made from retinal photographs.
Key Words
retinal photographs, magnification error, measurement,
image analysis
References
1. Blauth C, Arnold J, Kohner EM, Taylor KM. Retinal
microembolism during cardiopulmonary bypass demonstrated by fluorescein angiography. Lancet.
1986;ii:837-839.
2. Blauth C, Arnold JV, Schulenburg WE, McCartney A,
Taylor KM. Cerebral microembolism during cardiopulmonary bypass: retinal microvascular studies in
vivo using fluorescein angiography. J Thorac CardiovascSurg. 1988;95:668-676.
3. Blauth CI, Smith PLC, Arnold JV, Jagoe JR, Wootton
R, Taylor KM. Influence of oxygenator type on the
prevelence and extent of microembolic retinal ischaemia during cardiopulmonary bypass: assessment by
digital image analysis. J Thorac Cardiovasc Surg.
1990;99:61-69.
4. Jagoe JR, Blauth CI, Smith PL, et al. Quantification o_
retinal damage during cardiopulmonary bypass: comparison of computer and human assessment. Proc IEE.
1990; 137 Part 1(3):17O-175.
5. Jagoe J, Blauth C, Smith P, Arnold JV, Taylor KM,
Wootton R. Automatic geometrical registration of fluorescein retinal angiograms. Computers and Biomedical
Research. 1990; 23:403-409.
6. Arnold J, Jagoe R, Blauth C, Wootton R, Smith PLC,
Taylor KM. Computerised image analysis for the detection of capillary dropout in fluorescein angiograms. Invest Ophthal Vis Sci. 1990;31(suppl):138.
7. Arnold JV, Blauth CI, Smith PLC, Jagoe JR, Wootton
R, Taylor KM. Demonstration of cerebral microemboli occurring during coronary artery bypass graft
surgery usingfluoresceinangiography./ Audiov Media
Med. 1990;13(3):87-90.
8. Arnold JV, Jagoe R, Blauth C. The detection of microembolic lesions from the complete angiographic
sequence obtained during cardiopulmonary bypass
using digital image analysis. Invest Ophthalmol Vis Sci.
1991;32:866.
9. Baumbach P, Rassow B, Wesemann W. Absolute ocular fundus dimensions measured by multiple beam interference fringes. Invest Ophthalmol Vis Sci.
1989;30:2314-2319.
10. Bengtsson B, Krakau C. Some essential features of the
Zeiss fundus camera. Ada Opthalmol. 1977; 55:123131.
11. Littmann H. Zur bestimmung der wahren grosse eines
objektes auf dem hintergrund des lebenden auges.
Klin Monatbl Augenheilkd. 1982; 180:286-289.
12. Lotmar W. Dependence of magnification upon the
camera to eye distance in the Zeiss fundus camera.
Ada Ophthalmol. 1984;62:131-134.
13. Littman G, Summerer G. Fundus photography on Polaroid film. Zeiss Inf. 1970; 76:46-49.
14. Pach J, Pennel DO, Romano PE. Optic disc photogrammetry: magnification factors for eye position,
centration, and ametropias, refractive and axial; and
their application in the diagnosis of optic nerve hyperplasia. Ann Ophthalmol. 1989;21:454-462.
15. Snead MP, Rubinstein MP, Hardman LS, Havvorth
SM. Calculated verses A-scan result for axial length
using different types of ultrasound probe tip. Eye.
1990;4:7l8-722.
Downloaded From: http://iovs.arvojournals.org/pdfaccess.ashx?url=/data/journals/iovs/933174/ on 06/16/2017