RNFL Progression Analysis with the RTVue

Software Version 4.0

By

Michael J. Sinai, PhD

45531 Northport Loop West

Fremont, CA 94538
+1.510.623.8868
www.optovue.com

Copyright 2008 Optovue, Incorporated. All rights reserved. The material contained in this document may not be used, altered or edited in any
way without expressed written permission from Optovue, Inc. Part No. 300‐44062 Rev. A
 Glaucoma is a progressive optic neuropathy characterized by a loss of retinal ganglion cells,
thinning of the retinal nerve fiber layer (RNFL), thinning of the neuro‐retinal rim, and enlarged cupping.
Scanning laser devices like the OCT can be helpful to detect these changes. In fact, because of the high
level of reproducibility of the quantitative results from these scanning laser devices, small changes may
be detected more reliably with them than through direct clinical examination or with stereo‐
photographs. Several studies have found the accuracy for detecting glaucoma progression with optic‐
disc photographs is poor, mainly because the reproducibility of judging the discs is poor (Coleman et al.,
1996; Azuara‐Blanco et al. 2003). These structural changes can help determine not only whether a
patient is progressing or stable, but they can also help with the initial diagnosis as well. When
diagnosing glaucoma, imaging devices are helpful because they provide a quantitative assessment of
retinal structures (e.g., RNFL thickness) and can determine where a given patient stands in relation to a
pre‐defined normative range based on a normative database. The normative database is used to set
cut‐offs based on the statistical distribution of normal eyes, a patient’s result is then compared to this
normal distribution and when their value falls outside the normal range the software flags this result.
The cut‐offs used by the normative database are typically the bottom 5% of the normal distribution for a
classification of ‘Borderline’ and 1% for a classification of ‘Outside Normal Limits’. However the number
of ganglion cells in a normal eye can vary up to two‐fold, from approximately 750,000 to 1.5 million. So
the normal range can be very large, requiring a loss of a majority of ganglion cells before the value falls
outside the normal range for many patients. However progression analysis can potentially detect much
smaller changes and therefore the diagnosis of glaucoma can be made before a substantial loss of
ganglion cells occurs, even though the absolute value may still be within the normal range. The key
issue is the ability to detect significant change from a patient’s own baseline result.

In order to detect progression in glaucoma, the following three criteria must be met: 1) the
measurements must be reproducible, 2) the images must be accurately registered to each other, and 3)
a statistical test must be performed to differentiate true biological change from normal measurement
variability. Another important feature of meaningful progression analysis is to determine not only
whether significant change has occurred, but also the rate of change. Although both are important, the
rate of change may be more clinically meaningful because it can differentiate more stable patients from
those that are progressing more quickly. Data regarding rate of change can have direct clinical
relevance in regard to treatment strategies when related to the patient’s age and life expectancy.
Investigators have shown that a small, but significant number of patients may show rapid glaucomatous
changes (fast progressors), who can lose several decibels of vision each year (Artes 2008). These are the
patients who are likely to go blind if they are young at diagnosis and non‐aggressive therapy is used.
Conversely, a slowly progressing patient diagnosed at an older age may be treated less aggressively. The
rate of change in a patient is also helpful when determining how aggressively a patient should be
followed.

Copyright 2008 Optovue, Incorporated. All rights reserved. The material contained in this document may not be used, altered or edited in any
way without expressed written permission from Optovue, Inc. Part No. 300‐44062 Rev. A
Criteria 1: Reproducibility

Several independent investigators have shown the RTVue to have excellent reproducibility. For
reproducibility studies, most investigators report either the Coefficient of Variance (lower is better), or
the Intraclass correlation coefficient (higher is better). David Huang and his colleagues found the
Coefficient of Variance (CV) of the RTVue was 1.6% for normals, 1.7% for pre‐perimetric glaucoma
patients, and 2.1% for perimetric glaucoma patients (i.e., patients with reproducible visual field defects)
(Tan, Lu, Chopra, Varma, Ishikawa, Wollstein, Schuman, Huang, 2008). A study from UCSD also reported
excellent reproducibility, with the CV of the RTVue for average RNFL thickness to be 2.1%, and the
intraclass correlation coefficient was 0.97 (Gonzalez‐Garcia, Vizzeri, Bowd, Medeiros, Zangwill, Weinreb,
2008). In addition, studies have compared the RTVue reproducibility with other imaging instruments.
Aliyeva, Berisha, Roshdy, Pfeiffer, and Hoffmann (2008) found the inter‐observer agreement was
significantly better for the RTVue compared to the HRT, and Tan et al. (2008) found the RTVue had
better reproducibility than the Stratus (lower CV). One likely explanation for the improved
reproducibility compared to the Stratus comes from the fact that the Stratus takes a single circular scan
around the optic nerve head. The exact position of this circle can affect the RNFL results (see Figure 1.)

Figure 1. A comparison of
the scan circle location on
the RNFL thickness values.
In a normal eye, when the
scan is well centered, the
TSNIT will show a double‐
hump pattern (far left).
However if the scan circle is
not well centered and is too
low, then the RNFL thickness
in the TSNIT is increased
superiorly and decreased
inferiorly (middle pattern).
The opposite affect will
occur if the scan circle is too
high. This is caused by the
distance from the scan circle
to the disc margin.

If the operator does not have the scan perfectly centered over the optic nerve head (ONH), or if there
are eye movements during the scan, spurious RNFL thickness values can result which can make a normal
eye look glaucomatous and vice versa. For example in the middle panel of Figure 1, the scan circle is too
low relative to the center of the optic disc. This makes the superior part of the scan circle closer to the
disc, artificially increasing the RNFL thickness, and the inferior part farther from the disc, making the
RNFL artificially thinner.

Copyright 2008 Optovue, Incorporated. All rights reserved. The material contained in this document may not be used, altered or edited in any
way without expressed written permission from Optovue, Inc. Part No. 300‐44062 Rev. A
 The RTVue is not affected by this limitation because the RNFL map is made up of more data
points covering a large area around the ONH. The RNFL map covers a 5 mm diameter circle around the
ONH. The TSNIT graph is generated from a 3.45 mm diameter circle that is always centered on the ONH.
If the scan pattern is not perfectly centered on the ONH during the scan acquisition, the location of the
TSNIT circle can be adjusted left, right, up, or down, so that it is centered on the center of the ONH. This
is possible because of the large RNFL map that is collected, rather than a single circle. Because there is a
large RNFL map covering a large area, the exact location of the TSNIT circle can be adjusted so that it is
always centered on the optic disc. This is not possible with a single circle scan acquired in the Time
Domain Stratus. The time for the RTVue to collect the larger RNFL map is still less compared to the time
required for a single circle with the Time Domain OCT (due to a faster scan speed of 26,000 A scans/sec.
vs 400 A scans/sec with Time Domain OCT).

Criteria 2: Registration

A series of scans are automatically aligned and registered to each other based on matching the
blood vessel pattern from one scan to the next. Blood vessel based registration is likely the most
accurate method because adjustments can be made vertically, horizontally, and also rotationally in
order to precisely align one scan to the next. This ensures the same location is compared over time for
the most accurate change analysis possible.

The procedure for automatic image registration is as follows:

Step 1) A 3‐D scan of the optic disc is captured and the software automatically delineates the
optic disc margin. This is based on software algorithms that automatically detect the end of the RPE or
Bruch’s membrane at the disc margin (see Figure 2). This automatic detection is then displayed and the
operator can accept or modify the data points if necessary.

Figure 2. A display of the

result from the automatic optic
disc margin detection in a 3‐D
optic disc scan. The software
displays the results and allows
the operator to verify accuracy
and modify data points if
necessary. The disc margin
drawing is saved and is used
for all follow‐up exams.

Copyright 2008 Optovue, Incorporated. All rights reserved. The material contained in this document may not be used, altered or edited in any
way without expressed written permission from Optovue, Inc. Part No. 300‐44062 Rev. A
 Step 2) Next, the new ONH scan is taken in order to get the optic disc parameters and the RNFL
map. The new software automatically detects the RPE/Bruch’s membrane end and displays this result
to the user in order to verify accuracy (Figure 3). This screen display is similar to the old software
version, however the detection process is now automatic instead of manual.

Figure 3. The results of the

automatic detection of the
RPE/Bruch’s membrane end at
the disc margin. The software
displays the results and allows
the operator to verify accuracy
and modify data points if
necessary. This process is
performed after every ONH
scan.

Once the optic disc margin has been detected from the 3‐D scan, and the RNFL/Bruch’s membrane
endpoint have been detected in the ONH scan, the software places the disc margin circle over the ONH
scan results and calculates the optic disc parameters and RNFL map. In this way, an accurate drawing of
the optic disc margin is obtained, and the quantification of RNFL thickness and optic disc parameters
(e.g., cup area and rime area) is calculated. Figure 4 shows the RNFL thickness map and optic disc cup
and rim area superimposed on the 3‐D scan. The disc margin is outlined in red, and the light gray region
inside the disc is cup, and the dark gray region is rim. The RNFL thickness map is color coded where
thicker RNFL values are colored red and yellow and thinner are blue and green),

Copyright 2008 Optovue, Incorporated. All rights reserved. The material contained in this document may not be used, altered or edited in any
way without expressed written permission from Optovue, Inc. Part No. 300‐44062 Rev. A
 Figure 4. The result of registration
of the ONH scan to the 3‐D optic
disc scan. Notice the disc margin
drawing from the 3‐D scan is shown
and outlined in red. The cup and
rim areas are shown inside the disc
margin as light gray and dark gray
respectively. The RNFL thickness
map is shown in color with brighter
colors representing thicker areas
and darker colors representing
thinner areas.

For follow‐up exams, only the ONH scan needs to be taken, the 3‐D optic disc scan needs to be taken at
the initial visit only. The disc margin drawing from the 3‐D scan is automatically placed on all
subsequent ONH scans.

Step 3) The blood vessel map for each ONH scan is determined using the en‐face view of the
ONH scan. When viewed from the top‐down perspective, or en‐face, the blood vessel pattern around
the optic nerve is readily apparent, and can be detected and mapped automatically with software
algorithms. This mapping can be seen in Figure 5.

Figure 5. The Enface view of the

ONH scan reveals the blood vessel
pattern around the optic nerve
head. The 4.0 software
automatically detects this pattern
and registers all scans in a series to
each other based on this pattern.

Copyright 2008 Optovue, Incorporated. All rights reserved. The material contained in this document may not be used, altered or edited in any
way without expressed written permission from Optovue, Inc. Part No. 300‐44062 Rev. A
Criteria 3: Statistical Analysis

The progression analysis of the ONH scans employs a rigorous statistical analysis to a series of
scans in order to determine whether significant change has occurred and also to determine the rate of
change in microns/year. The statistical test used is called Statistic Image Mapping (SIM) adapted from
neuro‐imaging and applied to retinal imaging data by Dr. Garway‐Heath and colleagues at Moorfields
Eye Hospital (Patterson, Garway‐Heath, Strouthidis, Crabb, 2005). This technique has been found to be
more accurate for both sensitivity to detecting true change, and specificity for detecting no change in
stable eyes, than the Topographical Change Analysis (TCA) utilized in the Heidelberg Retinal
Topographer (HRT). The SIM method demonstrates a specificity of 90% vs 85% for TCA (specificity
means accurately detecting no change in stable eyes). Likewise the SIM method demonstrates a 73%
sensitivity for detecting change in ocular hypertensive converters compared to 53% for TCA (sensitivity
is the ability to detect change in eyes that converted to glaucoma). The authors conclude, “SIM has
better diagnostic precision in detecting change in a series of HRT images when compared to current
quantitative techniques” (Patterson, Garway‐Heath, Strouthidis, Crabb, 2005).

The statistical approach used in the SIM method is straightforward. First, a series of data points
is identified for analysis. These points can theoretically be anything, from a global parameter such as
average RNFL thickness, down to a single pixel. In the current software (version 4.0), there are 6 RNFL
sectors that are analyzed (Superior Temporal, Superior Nasal, Nasal, Inferior Nasal, Inferior Temporal,
and Temporal), as well as the global RNFL thickness parameter (see figure 3). After the images are
aligned based on the blood vessel registration algorithm, the data points at a given location are
arranged in order and subjected to a trend analysis. A regression line is created along with a standard
error based on the “goodness of fit of the regression line” to the actual data points (better fit means
smaller standard error). The slope is determined from the regression analysis. A simple test statistic is
created by taking the slope and dividing it by the standard error. If the slope is large, and the error is
small, the test statistic will be large and indicates likely progression. In the next step of the analysis, the
data points are randomly re‐arranged. After the random shuffling of the data, a new regression line
and slope are calculated from this new data re‐arrangement, along with a new error term. This random
permutation of the data and re‐calculation of the possible slopes and error terms is repeated 1,000
times in order to create 1,000 unique test statistics. Then a frequency distribution is created based on
all generated test statistics from this random permutation process (a histogram with 1,000 points).
Next, the typical cut‐off points are determined from this distribution, notably the 5% and 1% levels.
Finally the actual test‐statistic is compared to the distribution and cut‐off values. If the true test statistic
exceeds the level of the 5% cut‐off, it is colored yellow and labeled, “Possible Progression”, if the true
test statistic exceeds the level of the 1% cut‐off, it is colored red and labeled, “Likely Progression”,
otherwise it is colored green to indicate no significant progression was detected. The results are
presented in a new printout format showing the data points, regression line, slope, and total change for
each RNFL sector analyzed (See Figure 5).

Copyright 2008 Optovue, Incorporated. All rights reserved. The material contained in this document may not be used, altered or edited in any
way without expressed written permission from Optovue, Inc. Part No. 300‐44062 Rev. A
 Figure 5. The printout for the RNFL progression analysis. At the top the individual exam thickness
maps are presented. The baseline or initial exam is on the left, and each follow‐up is to the right.
The most recent follow‐up is presented in the middle of the printout. The trend analysis graphs for
each local region are shown surrounding the thickness map for the most recent follow‐up (middle).
The individual data points are shown in each graph, along with a regression line. The slope and total
change are adjacent to each graph and color coded based on the statistical analysis (see text for
details). At the bottom the TSNIT graph is shown with all exams plotted together. On the bottom
right is the trend analysis for the average RNFL thickness.

Copyright 2008 Optovue, Incorporated. All rights reserved. The material contained in this document may not be used, altered or edited in any
way without expressed written permission from Optovue, Inc. Part No. 300‐44062 Rev. A
Scan Pattern Details for Progression Analysis

The RNFL and optic nerve scan pattern used for the progression analysis has been slightly
modified in software version 4.0. The original RNFL and optic nerve scan pattern in older software
versions was called NHM4, and consisted of an RNFL map and optic disc analysis based on a scan pattern
with 12 radial lines and 6 concentric circles with a maximum diameter of 4 mm. The new scan pattern in
version 4.0 is now called ONH (for optic nerve head) and provides a larger scan area, out to 4.9 mm.
This creates a larger RNFL map and allows for more optic disc decentration than the older scan pattern.
This will be helpful for progression analysis where scans must be registered to each other and some area
of the RNFL map may not be included in all scans do to scan alignment. The details of the new ONH scan
pattern are as follows: Area scanned is 4.9 mm centered on the optic disc, scan pattern is made up of
13 concentric circles with diameter from 1.3mm to 4.9 mm with 0.3mm interval and 12 radial lines, total
data points is 14241, and the scan time is 0.55 seconds. The software will allow progression analysis to
be performed with both scan patterns, so a patient can be followed that started with the NHM4 scan
pattern, and then switched to the ONH scan pattern without any problem.

References

AL Coleman, A Sommer, C Enger, HL Knopf, RL Stamper, DS Minckler. Interobserver and intraobserver

variability in the detection of glaucomatous progression of the optic disc. J Glaucoma. 1996; 5:384‐9.

A Azuara‐Blanco, LJ Katz, GL Spaeth, SA Vernon, F Spencer, IM Lanzl. Clinical agreement amoung

glaucoma experts in the detection of glaucomatous changes of the optic disc using simultaqneous
stereoscopic photographs. Am J Ophthalmol. 2003; 136:949‐50.

PH Artes. Rates of visual field progression in treated and untreated glaucoma. OGS Annual meeting.
Anaheim CA 2008.

O Tan, AT Lu, V Chopra, R Varma, H Ishikawa, G Wollstein, JS Schuman, D Huang. Glaucoma diagnosis by
mapping peripapillary nerve fiber layer thickness with Fourier domain OCT. Invest Ophthalmol Vis Sci.
Suppl. 2008; 4655.

A.O. Gonzalez‐Garcia, G. Vizzeri, C. Bowd, F.A. Medeiros, L.M. Zangwill, R.N. Weinreb. Reproducibility
of RTVue Retinal Nerve Fiber Layer (RNFL) Thickness and Optic Disc Measurements and Agreement With
Stratus OCT Measurements. Invest Ophthalmol Vis Sci. Suppl. 2008; 4625.

AJ Patterson, DF Garway‐Heath, NG Strouthidis, DP Crabb. A New Statistical Approach for Quantifying

Change in Series of Retinal and Optic Nerve Head Topography Images. Invest Ophthalmol Vis Sci. 2005;
46:1659‐1667.

Copyright 2008 Optovue, Incorporated. All rights reserved. The material contained in this document may not be used, altered or edited in any
way without expressed written permission from Optovue, Inc. Part No. 300‐44062 Rev. A