Assessment of accuracy and repeatability of anterior segment optical coherence tomography and reproducibility of measurements using a customised software program


  • Eon Kim MBiomedE,

    Corresponding author
    1. Vision Cooperative Research Centre, Sydney, New South Wales, Australia
    2. School of Optometry and Vision Science, The University of New South Wales, Sydney, New South Wales, Australia
    3. Brien Holden Vision Institute, Sydney, New South Wales, Australia, E-mail:
    Search for more papers by this author
  • Klaus Ehrmann PhD

    1. Vision Cooperative Research Centre, Sydney, New South Wales, Australia
    2. School of Optometry and Vision Science, The University of New South Wales, Sydney, New South Wales, Australia
    3. Brien Holden Vision Institute, Sydney, New South Wales, Australia, E-mail:
    Search for more papers by this author

Eon Kim, Level 3, North Wing,, RMB, Gate 14,, Barker Street, The University of New South Wales, Sydney, NSW 2052, AUSTRALIA, E-mail:


Purpose:  The aim was to study the reliability of measurements of the RTVue (Optovue, Fremont, CA, USA) anterior segment optical coherence tomographer (AS-OCT) and assess how results can be improved by analysing raw optical coherence tomography data with customised image analysis software and applying correction factors.

Methods:  Five RTVue AS-OCT instruments (ver. 4.0) were assessed by imaging gauge blocks of three different lengths, single/stepped glass plate (microscope slides) and flat window glass to check for width, depth and linearity of the measurement scans. Five repeats per calibration tool were imaged and averaged. Raw data were exported and loaded into customised image analysis software written in LabWindows/CVI for further analysis. Using two calibration balls with different radii, measurement scans were validated. Repeatability of the optical coherence tomographs and the edge detection procedure were checked and statistical analyses performed.

Results:  Variations ranging from 0.01 to 1.93 mm in scan width and 0.1 to 0.17 mm in scan depth were found between the five instruments. Slight curvature distortion of 0.06 ± 0.01 mm (mean and standard deviation) was found in the raw images. By isolating the three sources of image distortion and applying individual correction factors, accuracy for corneal curvature measurements could be improved to better than 0.1 mm. Manual edge detection limited the coefficient of repeatability value to 0.06 and 0.08 mm for anterior and posterior radii of curvature, respectively. The coefficient of repeatability of corneal thickness measurements was less than 8 µm.

Conclusions:  Accuracy of the RTVue AS-OCT varied between instruments. By applying calibration scale factors calculated by customised software, accuracy of thickness and curvature values of the anterior eye was improved. The achievable precision is sufficient to detect clinically relevant corneal curvature variations.

Accurate in vivo measurements of the anterior eye are critical for keratorefractive and cataract surgery and to determine corneal disorders. Measurements of corneal thickness and curvatures are required for calculating the refractive power of the eye and form the basis for optical intervention or correction. Therefore it is crucial to calibrate and quantify instruments, which measure these parameters, to assure that they provide reliable data. According to the International Organisation for Standardization (ISO) standards (ISO 5725-1:19941 and ISO 5725-2:19942), accuracy refers to the closeness of agreement between the measurements and the true reference value. Repeatability and precision refer to the closeness of agreement between test results.1,2 Many previous studies have reported on the repeatability and precision of commercial ocular imaging instruments, from the traditional ultrasonic and optical pachymetry3 and corneal topography4,5 to the more sophisticated Scheimpflug imaging technology6–8 and optical coherence tomography (OCT).9–11

Optical coherence tomography is a non-invasive method, which can produce cross-sectional images of the anterior or posterior eye.12 While retinal OCT images are mainly used for qualitative evaluation, for the anterior segment images, quantitative presentation of thickness, angle and curvature data is of clinical interest and most instruments have some integrated image analytical tools to extract numerical parameters. There are distortions apparent in the OCT images, which have not been extensively studied previously.13–16 The fan distortion is caused by the scanning mechanism and the optics of the OCT system and Ortiz and colleagues13 proposed an algorithm to correct this distortion. We have previously17 presented a way to correct optical distortions by dividing the optical path of each layer by its corresponding refractive index. Ortiz and colleagues15 proposed application of a three-dimensional optical distortion correction, which increased the accuracy compared with prior methods. All distortions of the image affect the accuracy of the measurements obtained using OCT instruments. In the present study, the distortions apparent in the OCT images have been isolated into three individual causes: correcting for the scan width, depth and curvature distortion. The purpose of the present study is to assess the repeatability and accuracy of five RTVue OCT (Optovue, Fremont, CA, USA) instruments and to compare numerical results against reference standards and against results obtained from a customised software image analysis program. This will provide a baseline for the comparability of data obtained from each instrument and the basis to quantify and separate optical distortions for subsequent image correction.


Optical coherence tomography is a powerful imaging technique, which uses low coherence interferometry to generate cross-sectional two- or three-dimensional images of tissues or objects using a short coherent light beam. A Michelson interferometer splits the emitted light source into a sample and reference arm and then recombines the backscattered light from the sample arm with the light reflected off the reference arm. Interference will occur when the rays of light from these two arms have travelled the same optical distance. Recording the interference pattern along the depth range of the beam provides the A-scan and combining many A-scans of the laterally scanned beam generates the B-scan or two-dimensional cross-sectional image. There are two main types of OCT system: time-domain and frequency domain OCT.

The RTVue is based on the frequency domain OCT method, whereby the entire depth range (A-scan) is detected simultaneously using a spectrometer. This greatly increases the scanning rate compared with the time-domain OCT, which scans each depth one at a time. Many B-scan images can be combined to either obtain three-dimensional volume scans or to increase the signal-to-noise ratio by averaging B-scans of the same section. The RTVue OCT was originally designed to capture retinal two- and three-dimensional images. By placing either the cornea-anterior module long CAM-L (low magnification) or cornea-anterior module short CAM-S (high magnification) adaptor lens in front of the retina objective, the anterior segment of the eyeball can be imaged. The RTVue OCT uses 840 ± 10 nm wavelength. The measurement window can be set to a maximum 3.0 mm axial distance and 10 mm or 2.0 mm lateral for the long CAM-L and short CAM-S, respectively. Both lenses were tested in the present study but results are presented for CAM-L only because the image distortions with the CAM-S lens were much lower. The high axial resolution of 5.0 µm allows visualisation of different layers giving biometric measurements of the cornea.

Generating undistorted images of known scale factor requires either detailed knowledge of the A- and B-scan composition or retrospective image calibration based on scanning samples of known shape and refractive index. For the A-scan, assumptions have to be made on the refractive index of the sample to convert the optical path length to physical distances. Because the instrument has no way to identify individual layers with their various refractive indices, most anterior segment OCT assume a general refractive index value of 1.38, which provides only an approximation of the actual optical to physical thickness conversion. For the B-scan, the assumption is made that all A-scans are parallel, of even spacing and equi-distant from the reference mirror. For example, if any of these latter conditions is not met, the B-scan of a flat glass window would appear as a curved or tilted (or both) line (Figures 1A and 1B). By scanning objects of known shape and dimensions, the individual components of the distortion can be isolated, quantified and corrected. As a first step, a semi-automated edge detection process was implemented to extract contour outlines from the raw tomography data.

Figure 1.

Optical coherence tomography image of the flat glass window showing a slightly curved line representing a distortion on the image: (A) horizontal scan and (B) vertical scan.

Customised software image analysis program

This software image analysis program is customised to extract and compare accurate numerical parameters of the anterior eye. It is intended to provide more flexible calculations, such as applying a specific refractive index used for different layers of the cornea and applies a number of calibration factors to reduce scale and distortion errors.

The interference signals of the tomography sensor are stored as eight-bit intensity values in the form of rows and columns in a large binary format table. These data were exported and loaded into a customised LabWindows/CVI program (National Instruments, Austin, TX, USA), which converts the data into square pixels to form a raw image for further processing and analysis.

Edge detection

The region of interest of the image was selected manually by the user and using the ‘imaqSpoke’ function from National Instruments ‘nivision.h’ library, spokes were created, which radiate downward to detect the pixels on the anterior surface and radiate upward to detect the pixels on the posterior surface of the cornea. The edge detection algorithm searches for a sudden rise in pixel intensity to define a surface point. The inconsistent image quality, where average or regional intensity values of the pixels as well as random noise patterns might vary from image to image, made it necessary to include a manual edge detection intervention to allow users to specify individual edge points along the spokes. This enables more accurate results by identifying obvious outliers and either discarding or replacing them with points closer to the outline of the surface.

Circle fit

The detected edge points along the anterior and posterior surfaces were fitted with a circle fit, which is a default function provided in the ‘nivision.h’ library. The centre point and radii were determined.

Ellipse fit

Using the radius and centre point determined by the circle fit, an ellipse was fitted using the ellipse equation (1). The ‘a’ and ‘b’ values were varied iteratively by changing the radius of curvature (r0) and asphericity or shape factor ‘p’ value depending on how much the circle fit deviated from the raw detected data.


where ‘a’ is the major axis containing the r0 and ‘b’ is the minor axis and inline image, inline image.

Radius of curvature and the shape factor ‘p’ were determined by equation (2). For a sphere p = 1, flattening ellipse 0 < p < 1 and steepening ellipse p > 1.


Manual centre thickness

Using the two cursors from the customised program, thickness values were determined manually for the cornea. Thickness for different layers of the cornea can be calculated using their own specific refractive indices. The refractive index for the cornea used in the calculation was 1.370, which was extrapolated for wavelength 840 nm.18 The same edge detection and circle fit tools and methods were also used for the OCT images obtained from the calibration balls, gauge blocks and glass plates as detailed below.

Edge detection repeatability

Three subjects were used to check the repeatability of the edge detection method. Five repeat measurements of each eye were taken using one OCT instrument. Two independent images of a cornea were randomly chosen from the collection of images between the three subjects. The subjects were all consenting researchers with no known ocular pathology; their details are listed in Table 1. Edges of anterior and posterior corneal surfaces were selected by a single operator using the manual edge detection intervention for each of the two images. Radius of curvatures and shape factors were calculated. The operator was not masked to previous measurements; however, this process was repeated for each image over five different days to avoid bias measurements and to determine the 95% confidence interval of the edge detection method.

Table 1. Description of the three subjects participating in the present study. Sph/Cyl = Spherical/Cylinder
 Refractive errorAge (years)Gender
Sph/Cyl * axis
Subject #1+0.25/+0.50*1036M
Subject #2+0.25/-0.50*548F
Subject #3+2.75/-0.25*12846F

Accuracy of the instrument

The three main parameters which determine the accuracy of the instrument are: scan width, scan depth and curvature distortions. Each of these error sources were analysed and corrected independently with the use of suitable calibration normals. Images of gauge blocks (Mitutoyo Gauge Block Set, Grade 1; Kawasaki, Japan) of three different lengths (4.0, 6.0 and 8.0 mm), a flat glass window (CVI Melles Griot, Albuquerque, NM, USA; 02WBK005 BK7), a stepped glass plate (Lomb Scientific, Scoresby, VIC, Australia, BK7 microscope slides, refractive index: 1.5102) and two ceramic calibration balls (Saphir Werk Industrie Produckte AG, Bruegg, Switzerland) of different radii, 7.0 mm (14-00-070-KE, Grade 10) and 9.0 mm (18-00-072-KE, Grade 5), were taken with five different RTVue OCT instruments. The manufacturer's accuracy was better than 100 nm in all cases. First, using the built-in software calliper tools from the OCT (ver. 4.0), scan widths and depths were checked for accuracy. Then, using the raw data of the images loaded onto the customised software image analysis program, scan widths and depths were checked again and the calibration factors were determined for each of the five instruments.

Scan widths were checked using the gauge blocks of different lengths. The number of pixels determining the length of the gauge block was obtained using the two cursors on the program and converted to millimetres using equation (3) with the given scan width and total number of x pixels of the image.


The scale factor for scan width was calculated using a linear fit, real value length against the calculated length of the gauge block.

Scan depths were checked using the thickness of the single glass plate and the step height of a stepped glass plate (Figure 2). The physical thickness of the glass slide was first measured using a thickness gauge and the optical thickness was calculated using equation (4). The refractive index of the glass slide (BK7) at wavelength 830 nm, as used in the calculation, was 1.5102.

Figure 2.

Stepped glass plate made of two microscope slides bonded on top of another slide with a narrow gap between them: (A) image taken with the optical coherence tomography pupil camera; stepped glass plate labelled with ‘L’ and ‘R’ representing the left and right sides of the gap, respectively, across which the measurement is taken. (B) Optical coherence tomography image of the horizontal scan across the stepped glass plate showing the step height. (C) Schematic diagram (not to scale) of the stepped glass also showing the step height.


Using the same principle as equation (3), the new scan depth value was calculated with the thickness of the glass slide in millimetres, the number of pixels of the thickness of the glass slide and the total number of y pixels of the image.


Curvature distortion within the image was checked using the flat glass window. Any deviation from a straight line of the scanned window surface was quantified and stored for image correction.

Validation of all determined correction factors, used to scale and undistort the images, was performed by scanning the spherical calibration balls and comparing their corrected images with the nominal shape.

Measurement repeatability

Five images from each of the three right eyes and two calibration balls were taken by a single operator from one of the RTVue OCT instruments. Between the five images, the subjects were instructed to blink a couple of times. The corneal radius of curvature (r0) within the central 5.0 to 8.0 mm range was calculated. The shape factor ‘p’ was also calculated.

Statistical analysis

The coefficient of repeatability was calculated according to the Bland and Altman19 limits of agreement, which is defined as 1.96 times the standard deviation of differences between pairs of measurements. For edge detection repeatability, the standard deviation of the mean between two independent images repeated five times was used for the calculation. For measurement repeatability, the standard deviation of the mean was used for the calculation instead of the standard deviation of differences between pairs of measurements.


Customised software image analysis program

A typical OCT image of a human cornea loaded into the customised software image analysis program is shown in Figure 3. The red points along the vertical spoke lines signify the detected edge points on the front surface of the cornea, while the yellow points are those from the posterior cornea. Obvious outliers were replaced or removed manually.

Figure 3.

Screenshot optical coherence tomography image of a cornea loaded onto the customised software analysis program showing the detected cornea's edges along the spokes of each surface

Edge detection repeatability

Anterior and posterior surfaces ellipse fit radius of curvature (mean and confidence interval (mm)) for five repeated edge detection processes were 7.54 ± 0.02 and 6.58 ± 0.04, respectively, for Image 1 and 7.84 ± 0.02 and 7.08 ± 0.03, respectively, for Image 2.

Accuracy of the instrument

The result for the scan width, scan depth and calibration ball measurements from the five RTVue instruments are presented in Table 2. Results are given before and after applying the correction factors. For two instruments, width and depth measurements were also taken using the built-in callipers in the RTVue supplied analysis software. An example of how the OCT measurements were obtained using the built-in measurement tool is illustrated in Figure 4. The screenshot shows the OCT image of the 6.0 mm gauge block together with the distance measurement line as placed with the mouse cursor.

Table 2. Results (in mm) before and after applying the correction factors for five RTVue instruments using the raw data of the images. For RTVue #1 and RTVue #3, measurements were also obtained using the built-in callipers in the instruments' software.
 RTVue #1RTVue #2RTVue #3RTVue #4RTVue #5
 BeforeAfterBeforeAfter BeforeAfterBeforeAfterBeforeAfter
  • † 

    Scan depth before correction is in medium defined by the manufacturer and scan depth after correction is corrected for air.

  • OCT: optical coherence tomography

8 mm block9.859.887.957.977.988.017.888.017.947.958.027.98
6 mm block7.57.565.996.
Scan depth1.961.962.822.162.871.941.942.811.992.961.972.96
Glass slide thickness, 1.045 mm1.140.731.050.791.051.130.721.050.681.050.671.05
Step height0.721.040.701.020.691.06
Ball 7 mm16.006.999.456.9810.517.0110.367.0110.296.98
Figure 4.

Optical coherence tomography image of a 6.0 mm gauge block with its lengths determined using the calliper tools function from the built-in RTVue #1 instrument

Scan depth data were calculated using equation 5 mentioned in the Methods section with square pixels and linearity factors included in the calculations. For the glass slide thickness data, there are differences between the OCT and the uncorrected measures compared with the fully corrected results. This is largely due to the assumed refractive index of the cornea for the scanned object, whereas the actual measured distance was either in air or in glass. For the RTVue #1 instrument, there was also a significant error for the scan width. The correction factors of the scan widths and scan depths are shown in Table 3.

Table 3. Correction factors used for correcting scan widths and scan depths
 RTVue #1RTVue #2RTVue #3RTVue #4RTVue #5
Correction factor for new width0.800.991.011.001.00
Correction factor for new depth in air1.050.961.071.071.06

The corrected curvature values for the calibration balls take into consideration the correction factors for the scan width and depth as well as the curvature distortions within the image. The curvature distortions apparent in the images (Figure 1) are presented in Table 4.

Table 4. Curvature distortions in the images for each optical coherence tomography instrument, as determined by scanning a flat glass plate
 RTVue #1RTVue #2RTVue #3RTVue #4RTVue #5
Curvature error (mm)

The mean curvature values for the calibration ball from the five RTVue instruments, after applying the calibration scale factors, are shown in Figure 5 as the difference from the true value of the 7.0 mm ball. The horizontal lines indicate the confidence interval limits for each of the instruments. The fitted ellipses for the calibration balls were very close to the actual radii with a remaining maximum error of 0.018 mm. The result for the 9.0 mm ball was very similar to the 7.0 mm ball with a maximum error of 0.028 mm.

Figure 5.

Results of the mean radius of curvature minus the nominal value of the 7.0 mm calibration ball from five different RTVue instruments. Each set of horizontal lines matches the confidence interval limits for each of the RTVue instruments.

Measurement repeatability

The mean radius of curvature of five sets of independent measurements for both anterior and posterior surfaces of ellipse fit (r0) and shape factor ‘p’ are shown in Table 5 for three subjects and two calibration balls. The 95% confidence intervals of r0 and shape factor ‘p’ values are shown in Figures 6 and 7, respectively. These values were calculated using data points within two different central diameters (8.0 mm and 5.0 mm) of the cornea or balls. The confidence intervals for the central curvature r0 were all below 0.09 mm.

Table 5. Results of mean radius of curvature (r0) values and shape factor ‘p’ for both anterior and posterior surfaces
 Subject 1Subject 2Subject 3Ball 1 (7 mm)Ball 2 (9 mm)
r0 (mm)Shape factor ‘p’r0 (mm)Shape factor ‘p’r0 (mm)Shape factor ‘p’r0 (mm)Shape factor ‘p’r0 (mm)Shape factor ‘p’r0 (mm)Shape factor ‘p’r0 (mm)Shape factor ‘p’r0 (mm)Shape factor ‘p’
8 mm range7.510.956.530.837.941.097.060.957.750.836.910.986.991.108.991.00
5 mm range7.500.736.470.877.940.726.980.927.851.066.890.857.011.178.940.63
Figure 6.

Confidence interval values for anterior and posterior radii of curvature

Figure 7.

Confidence interval values for anterior and posterior shape factor ‘p’

The mean and standard deviation of the central corneal thickness for the three subjects are 575.8 ± 3.5, 597.7 ± 5.4 and 547.4 ± 4.6 µm. These values were obtained by manually selecting surface points of the central anterior and posterior cornea and dividing the optical thickness by the corneal refractive index. The standard deviation values ranged between 3.5 to 5.4 µm for the five repeated measures.

Statistical analysis of the measurement

The coefficient of repeatability was calculated for edge detection repeatability using the five repeated sets of two of the subjects and measurement repeatability using the images of the three subjects, separately for anterior and posterior curvature and central corneal thickness. These results are shown in Table 6. The overall measurement coefficient of repeatability was better for the anterior central curvature (0.11 mm) than for the posterior (0.16 mm). For the centre thickness, the coefficient of repeatability was 8.8 µm.

Table 6. Coefficient of repeatability (COR) calculated using Bland and Altman limits of agreement defined by 1.96 × SD of difference for edge detection repeatability and measurement repeatability
COREdge detectionMeasurement
Ellipse r0 ant0.04 mm0.11 mm
Ellipse r0 post0.08 mm0.16 mm
Centre thickness6.9 µm8.8 µm


A customised software image analysis program was written to use the raw data exported from the RTVue OCT and provide more flexible and accurate extraction of numerical data on the parameters. The RTVue comes with basic image analysis tools but these were rudimentary and, as shown in the present study, not very accurate. Dunne, Davies and Wolffsohn16 reported similar issues with the Visante AS-OCT (Carl Zeiss Meditec, Dublin, CA, USA) and presented a new scheme for measuring axial distances and surface curvature using their own customised image analysis program, which increased the accuracy of the results.

The process of repeatedly selecting the edges from a single image was to assess the error obtained from the manual edge detection method. This manual detection provides more reliable data compared with the automated edge detection, particularly when images are of poor quality and points had been detected, which did not fall on the corneal outline. Repeatability of the measurements was assessed using the five images of the three subjects. The coefficient of repeatability for edge detection was slightly lower compared with that obtained for the measurement repeatability as shown in Table 6. This is expected, as the errors from the edge detection process are only a subset of the total variability between independent measurements. Douthwaite and Parkinson4 reported the repeatability coefficient using the Orbscan II instrument for corneal curvatures. For a mean of five measurements, a value of 0.08 mm was reported compared with 0.11 mm determined in the present study.

Accuracy of the raw data varied among the five different RTVue OCT examined in the present study. Without further calibration, results obtained from different instruments would not be comparable. A small error in our calibration process was discovered after the data had been analysed. The step height of the glass plate imaged with RTVue #4 suffered an unintentional error due to the glue thickness, which was used to bond the glass plates together. The average thickness of the glue measured was 0.02 mm or approximately two per cent of the actual step height; an error that affects the total error insignificantly.

The calibration scale factor was determined for both horizontal and vertical scans of the cross-line scan type for each of the instruments and was applied to increase accuracy levels sufficient for all clinical applications. There was no significant difference in the scale factors between the horizontal and vertical scan among the same instruments. Images using the CAM-S lens ‘cross line’ scan were also checked and only the scan depth required scale factors to be applied to the raw data for accurate measurements for all five instruments. As illustrated in Figure 5, the accuracy was shown as the difference in the mean and the nominal values of anterior curvature of the calibration ball. The difference values were within the 95% confidence interval for three of the RTVues. Similarly, Perez and colleagues5 calculated the mean deviations from the nominal value using a steel ball, which were all less than ±0.05 mm.

The spherical calibration balls, Ball 1 and Ball 2 with radii of 7.0 and 9.0 mm, respectively, were used to verify the absolute accuracy of the calibrated OCT images. The measured radii of curvature of 6.99 ± 0.03 mm and 8.99 ± 0.05 mm from one randomly selected RTVue OCT instrument are well within clinically relevant error margins. The mean shape factors ‘p’ were 1.10 ± 0.15 and 1.00 ± 0.34 for Ball 1 and Ball 2, respectively. While the mean shape factor ‘p’ was close to the expected value of 1 for both of the balls, the shape factor ‘p’ was the most inconsistent of the measurements, mainly because the curve-fitting algorithm for extracting the shape factor ‘p’ is very sensitive to even small variations in sag of peripheral data points. Table 7 gives some examples of how little peripheral sag error would be required to cause the false shape factor results for the calibration balls. The results for the human corneal curvatures and shape factor ‘p’ for the three subjects fall within the average range as previously reported.20 As expected, fitting the conic section over a larger diameter provides better confidence intervals and generally a slightly flatter curvature.

Table 7. Results (in mm) showing the sag difference converted from the shape factor ‘p’ error margin
 Ball 1Ball 2
Image 1Image 2Image 1Image 2
8 mm0.01150.01360.00930.0070
5 mm0.00200.00200.00160.0010

Sin and Simpson21 also calculated the worst case coefficient of repeatability for corneal thickness, which was ±9.98 µm, and the intersession coefficient of repeatability was ±10.64 µm. The coefficient of repeatability for edge detection intervention and measurement reported in this study were slightly better (Table 6). Other studies compared the repeated measurement collected at different sessions. Muscat and colleagues11 studied the repeatability of the corneal thickness measurements using the Humphrey Zeiss OCT and reported 95% limits of agreement of -16 to +7 µm from the intersessional data, while Marsich and Bullimore3 found repeatability of the Orbscan system with 95% limits of agreement of -10 to +17 µm for day-to-day comparisons.

The accuracy of RTVues can vary significantly between instruments but can be improved by isolating individual error components and applying correction factors. The current study presented extensive calibration work that required accessing and loading the raw data into the customised image analysis software and applying calibration scale factors, as established for each individual instrument. There are some simple steps that can be implemented without the use of specialised hard or software tools to check and correct the scan length and scan width.

For scan width

Using the optical coherence tomographer, take a horizontal and vertical scan image of a ruler or object of known length.

Measure the distance in the OCT image of the known length of the object with the built-in calliper from the instrument software.

Calculate the scale factor by dividing the actual known length by the measured length (if more than one known length calibration tool is measured then fit a linear fit to calculate the scale factor).

For scan depth

Using the optical coherence tomographer, take an image of a single microscope slide glass (BK7) with known thickness.

Using the built-in calliper from the instrument software, measure the thickness of the microscope.

From the measured and actual thickness values of the glass slide, the correction factor (cf) for corneal thickness measurements can be determined using the following equation:


Whence, the 1.10 factor is the ratio of the glass and corneal refractive indices.

Once the correction factor is established for a particular instrument, all corneal thickness measurements obtained with the calliper should be multiplied by this correction factor to obtain the correct thickness value.

The present study focused on the anterior segment images using the OCT with the CAM-L adaptor lens. Absolute accuracy is clinically important when assessing corneal shape and thickness. The established correction factors cannot be applied directly when the instrument is used for retinal imaging, as the imaging system and the optics of the eye are very different. The only useful improvement in image accuracy can be achieved by applying different refractive indices to the different retinal layers to obtain more accurate layer thickness data.

One of the major advantages of the customised image analysis program presented in the present study is the potential to identify different layers and use different refractive indices within the images such as tears, contact lens, epithelium, stroma and aqueous and to apply individual factors for optical to physical thickness conversion. This is work in progress and will be presented in future publications. The extra effort of manual edge detection is justified for research and critical clinical applications where accuracy and reliability of corneal measurements are essential.


The authors gratefully acknowledge support from the Australian government through the Cooperative Research Centres scheme and The University of New South Wales Australian Postgraduate Award scholarship.