- Email: [email protected]
Quantitative Thickness Measurement of Retinal Layers Imaged by Optical Coherence Tomography
Quantitative Thickness Measurement of Retinal Layers Imaged by Optical Coherence Tomography MAHNAZ SHAHIDI, PHD, ZHANGWEI WANG, MS, AND RUTH ZELKHA, MS
● PURPOSE:
To report an image analysis algorithm that was developed to provide quantitative thickness measurement of retinal layers on optical coherence tomography (OCT) images. ● DESIGN: Prospective cross-sectional study. ● METHODS: Imaging was performed with an OCT3 commercial instrument in 10 visually normal healthy subjects. A dedicated software algorithm was developed to process the raw OCT images and detect the depth location of peaks from intensity profiles. Quantitative thickness measurements of three retinal layers, in addition to total retinal thickness, were derived. Total retinal thickness measurements obtained by the algorithm were compared with measurements provided by the standard OCT3 software. ● RESULTS: The total retinal thickness profile demonstrated foveal depression, corresponding to normal anatomy, with a thickness range of 160 to 291 m. Retinal thickness measured by the algorithm and by the standard OCT3 software were highly correlated (R ⴝ 0.98). The inner retinal thickness profile predictably demonstrated a minimum thickness at the fovea, ranging between 58 to 217 m along the 6-mm scan. The outer retinal thickness profile displayed a maximum thickness at the fovea, ranging between 66 to 107 m along the 6-mm scan. The photoreceptor outer segment thickness profile was relatively constant along the 6-mm scan through the fovea, ranging between 42 to 50 m. The intrasubject variabilities of the inner retina, outer retina, and photoreceptor outer segment thickness was 14, 10, and 6 m, respectively. ● CONCLUSIONS: Thickness measurements of retinal layers derived from OCT images have potential value for objectively documenting disease-related retinal thickness Accepted for publication Jan 6, 2005. From the Department of Ophthalmology and Visual Sciences, University of Illinois at Chicago, Chicago, Illinois. This study was supported by the Department of Veterans Affairs (Washington, DC); National Eye Institute EY14275 and EY01792 (Bethesda, MD); and an unrestricted fund from Research to Prevent Blindness (New York, NY). Inquiries to Mahnaz Shahidi, PHD, Department of Ophthalmology and Visual Sciences, University of Illinois at Chicago, 1855 West Taylor Street, Chicago IL 60612; fax: (312) 413-7366; e-mail: [email protected]
1056
©
2005 BY
abnormalities and monitoring progressive changes over time. (Am J Ophthalmol 2005;139:1056 –1061. © 2005 by Elsevier Inc. All rights reserved.)
I
MAGING AND QUANTITATIVE ASSESSMENT OF THE RET-
inal tissue is of value for the diagnosis of many retinal diseases. Optical coherence tomography (OCT) is an imaging technique that generates high-resolution crosssectional images of the retina and provides quantitative measurement of the total retinal thickness.1,2 OCT has been extensively applied for evaluating changes in retinal thickness resulting from various retinal diseases.3–7 Image analysis software that is provided by the commercially available OCT instrument allows measurement of total retinal thickness. Although intraretinal structures are visualized on OCT images, quantitative assessment of these structures is not obtainable by the instrument software. Recently, a new generation of OCT has been developed that has a higher depth resolution than the commercially available instrument8,9 and has been used to compare sublayers of the retina with histologic studies in the same eyes.10,11 This technology is in developmental stage, however, and not widely available because of its excessive cost. In this study, an image analysis algorithm was developed to measure the thickness of retinal layers from images obtained with the use of an OCT commercial instrument. Normal thickness measurements and intra- and intersubject variability were established from OCT images acquired in visually normal healthy subjects.
METHODS OCT IMAGING WAS PERFORMED USING AN OCT3 COMMER-
cial instrument (Carl-Zeiss Meditec, Dublin, California, USA). Six radial OCT scans, each 6 mm in length, were acquired. Each radial OCT scan was 1,024 pixels (2 mm) in depth and 512 pixels (6 mm) in length and thus had a depth resolution of 2 m/pixel and spatial resolution of 12 m/pixel. The vertical (scan 1 of 6) and horizontal (scan 4 of 6) OCT scans were analyzed. The raw gray scale OCT
ELSEVIER INC. ALL
RIGHTS RESERVED.
0002-9394/05/$30.00 doi:10.1016/j.ajo.2005.01.012
FIGURE 1. Outline of algorithm developed for analysis of optical coherence tomography (OCT) images. (A) OCT image in bitmap format obtained from a horizontal scan through the fovea obtained in one of the subjects in the study. (B) Exported OCT image in raw format. (C) Edge-enhanced processed image. (D) Location of the four peaks as identified on the edge-enhanced image: upper left, peak 1; upper right, peak 2; lower left, peak 3; lower right, peak 4. (E) Profiles of the identified peaks superimposed on the raw OCT image.
images were exported for analysis. A dedicated software program was developed in Matlab (Mathworks, Natick, Massachusetts, USA) for image analysis. The image analysis algorithm is outlined in Figure 1. A Gaussian function with normal distribution was generated and convolved with the raw image to derive an edge enhanced image. The image was cropped into a subimage of 15 (length) ⫻ 1,024 (depth) pixels. The subimage was averaged to derive an intensity profile. The profile was smoothed by median filtering and normalized. The points of maxima (peaks) and minima (valleys) of the profile were automatically identified. The first maximummost anterior peak of the intensity profile (peak 1) that had a value greater than 3 times the standard deviation of all peak intensities was identified, corresponding to the vitreoretinal interace. The second maximum peak, posterior to peak 1, was identified as peak 2, corresponding with the interface of inner and outer photoreceptor segments. The minimum valley between peaks 1 and 2, immediately anterior to peak 2, and with a value greater than three times the standard deviation of all valleys was identified as peak 3. The center between the 2 minimum valleys immediately posterior to peak 2 was identified as peak 4, corresponding to the chorioretinal interface. This procedure was repeated, each time shifting by 15 pixels (180 VOL. 139, NO. 6
m) along the image and generating another subimage. In this way, 34 coordinates (depth positions) for each of the four peaks were recorded along the 6-mm scan. Thickness profiles for three retinal bands were derived, in addition to the total retinal thickness, by calculating the peak separations and converting the pixel values to absolute thickness in microns. The separation between peak 1 (vitreoretinal interface) and peak 2 (inner and outer photoreceptor segments interface) was calculated along the 6-mm scan to provide a profile of the total retina (TR) thickness. The separation between peaks 2 and 3 provided a thickness profile of a retinal band designated as outer retina (OR) in this study. The separation between peaks 2 and 4 (chorioretinal interface) was calculated to provide a thickness profile of the photoreceptor outer segment (OS) layer. The separation between peaks 1 and 3 provided a thickness profile of the inner retina (IR). In each subject, OCT images from three vertical and three horizontal scans were analyzed, and the data were averaged to provide thickness profiles, along vertical and horizontal scans, respectively. Thickness profiles for each retinal layer, along the vertical and horizontal scans, were derived by averaging the profiles across all subjects.
QUANTITATIVE THICKNESS
OF
RETINAL LAYERS
1057
FIGURE 2. Orientations of the vertical and horizontal 6-mm OCT scans on a fundus image.
Intrasubject variability (reproducibility) was determined as the coefficient of variation (COV; standard deviation divided by the mean thickness) of three measurements in each subject, averaged over all subjects. Intersubject variability was determined as the standard deviation of thickness measurements across all subjects. OCT images were also analyzed using the standard OCT3 software. The TR thickness measurements obtained from three images in each eye were averaged. A TR thickness profile was generated by averaging the profiles in all subjects. The TR thickness profiles obtained by the Matlab algorithm and the standard OCT3 software were compared by linear regression analysis. Approval for use of human subjects was obtained from the Institutional Review Board of the University of Illinois at Chicago. The study was explained to the subjects, and informed consent was obtained according to the tenets of the Declaration of Helsinki. OCT imaging was performed in one eye of 10 visually normal subjects, 7 women and 3 men, five left and five right eyes. The subjects’ ages ranged from 25 to 56 years, with an average of 39 ⫾ 12 years (mean ⫾ SD).
FIGURE 3. Identification of retinal layers on an OCT image. The thickness of the total retina (TR), inner retina (IR), outer retina (OR), and photoreceptor outer segment (OS) are visualized on the OCT image. The outer retina (dark band) comprised outer nuclear cells and photoreceptor inner segments.
normal eye (Figure 2). The retinal layers as imaged by OCT are shown in Figure 3. The bands corresponding to IR, OR, OS, and TR are marked. The average TR thickness profiles derived from analysis of the vertical and horizontal scans in all subjects are shown in Figure 4. The average TR thickness profile demonstrated a foveal depression, corresponding to normal anatomy, with a thickness range of 160 to 291 m along the 6-mm scan. Mean retinal
RESULTS THE ORIENTATIONS OF THE VERTICAL AND HORIZONTAL
OCT scans are displayed on a fundus image of a visually 1058
AMERICAN JOURNAL
OF
OPHTHALMOLOGY
JUNE 2005
FIGURE 5. The relationship between total retinal thickness measured by the Matlab algorithm and by standard OCT3 software. Thickness measurements obtained by the two methods were highly correlated (R ⴝ 0.98).
FIGURE 4. Total retinal thickness profiles along the horizontal (top) and vertical (bottom) scans, averaged over all subjects and derived by the developed Matlab algorithm, corresponds to normal retinal anatomy. The standard deviation of measurements is indicated by error bars.
thickness was 243 ⫾ 35 m for the 6-mm vertical scans and 247 ⫾ 34 m for the horizontal scans. The mean intrasubject variability of total retinal thickness measurements derived by the COV was found to be 0.04, corresponding to a standard deviation of 10 m. The relationship between TR thickness profiles obtained by the Matlab algorithm and the standard OCT3 software is shown in Figure 5. The TR thickness profile mean obtained by Matlab algorithm and standard OCT3 software were 243 ⫾ 35 m and 232 ⫾ 35 m, respectively. The thickness measurements obtained by the two methods were highly correlated (R ⫽ 0.98). The thickness profiles of three retinal bands measured along the 6-mm OCT scan, averaged over all subjects, are shown in Figure 6. From the vertical scans, the mean thickness of profiles for the IR, OR, and OS retinal bands was 170 ⫾ 39 m, 78 ⫾ 10 m, and 47 ⫾ 2 m, respectively. From the horizontal scans, the mean thickness of profiles for the IR, OR, and OS retinal bands was 161 ⫾ 42 m, 81 ⫾ 13 m, and 46 ⫾ 2 m, respectively. The IR thickness profile demonstrated a minimum thickness at the fovea, ranging from 58 to 217 m along the 6-mm scan. The OR thickness profile displayed a maximum thickness of at the fovea, which ranged between 66 to 107 m along the 6-mm scan. The OS thickness profile was relatively constant along the 6-mm scan through the fovea, ranging between 42 to 50 m along the 6-mm scan. VOL. 139, NO. 6
FIGURE 6. Inner retina (IR), outer retina (OR), and photoreceptor outer segment (OS) thickness profiles along the horizontal (top) and vertical (bottom) scans, averaged over all subjects. The standard deviation of measurements is indicated by error bars.
The mean intrasubject thickness measurement variability (COV) of the IR, OR, and OS retinal bands on the vertical scans was found to be 0.08 (SD ⫽ 14 m), 0.13 (SD ⫽ 10 m), and 0.13 (SD ⫽ 6 m), respectively. The mean intrasubject thickness measurement variability (COV) of the IR, OR, and OS retinal bands on the
QUANTITATIVE THICKNESS
OF
RETINAL LAYERS
1059
horizontal scans was 0.09 (SD ⫽ 14 m), 0.15 (SD ⫽ 12 m), and 0.13 (SD ⫽ 6 m), respectively. The mean intersubject thickness measurement variability (COV) of the IR, OR, and OS retinal bands was found to be 0.11 (SD ⫽ 18 m), 0.14 (SD ⫽ 11 m), and 0.11 (SD ⫽ 5 m), respectively.
this band are ganglion cell axons, ganglion cell nuclei, inner plexiform, inner nuclear, and Henle fibers. Anatomically, the thickness of this band is expected to minimize at the foveal center because the thickness of most of these retinal layers becomes negligible at this point but maximizes within 1 mm nasal to the foveal center on the horizontal scan due to an increase in the ganglion cell axon and nerve fiber layer thicknesses in this region, consistent with the findings of our study. A dark band was observed, which signified the outer retina, as termed in this study. This retinal band comprised outer nuclear cells and photoreceptor inner segments. Our results indicate increased thickness of this band at the fovea as expected anatomically, partially due to the elongated foveal cone photoreceptors. A second dark band was identified bounded by chorioretinal interface and an intensity peak immediately anterior to the interface, which signified the region of photoreceptor outer segments.10 The intensity peak anterior to the chorioretinal interface on OCT images is likely to originate from the junction of the inner and outer photoreceptor segments.16 The thickness of this band was found to be relatively constant, consistent with an anatomically uniform thickness. In retinal diseases in which the photoreceptor cells have degenerated, a decrease in the thickness of outer retinal layers is anticipated. The reproducibility of outer retinal thickness measurements in normal subjects was found to be 10 m, indicating that thickness changes of 20 m can be detected over time with 95% confidence. The intersubject variability of the measurements was found to be 11 m, indicating that thickness abnormalities of 22 m can be detected with 95% confidence. In certain retinal degenerative diseases, the inner retinal thickness may also be altered. Based on the intra- and intersubject variability of the inner retinal thickness measurements, changes of 28 to 36 m can be detected with 95% confidence. The reproducibility of thickness measurements in patients with retinal pathologies or media opacities may be lower and needs to be established, however. The application of the described image analysis algorithm is limited to retinal pathologies that result in a relatively uniform thickness change in the retinal layers. Inhomogeneous retinal pathologies may cause local changes in the optical properties of the retinal layers. The influence of these changes on the peak detection algorithm and thickness measurements needs further investigation. Overall, thickness measurement of retinal layers has potential value for objectively documenting disease-related changes in retinal thickness and monitoring progressive changes over time.
DISCUSSION THE USEFULNESS OF OCT FOR IMAGING OF RETINAL PATHO-
logic changes in various diseases and detection of retinal thickness abnormalities has been established.6,7,9,12–15 The standard OCT3 software only offers measurements of total retinal thickness, however, and thus is limited in providing quantitative information on intraretinal structures. Development of new image analysis software that can measure the thickness of intraretinal structures provides additional information for documenting disease-related changes in the retina and has potential value for the diagnosis and monitoring of diseases in which depth-specific thickness abnormalities may be observed. In our study, development of an image analysis algorithm is reported that overcomes the limitation of the commercial OCT3 software and provides additional quantitative information that can be extracted from OCT images. The Matlab image analysis algorithm consists of processing of OCT images and detection of the depth location of peaks corresponding to high- and low-intensity bands. This algorithm can be applied to OCT macular images scanned at high resolution (1,024 ⫻ 512 pixels) by commercial OCT3 instrument. On the OCT images in normal eyes, three distinct retinal bands were identified. The thickness of these bands, in addition to the total retinal thickness, was measured. The intensity variation on OCT images stems from a combination of laser reflection, scatter, and absorption. At the vitreoretinal interface, a bright signal (peak in intensity profile) was detected because of a change in the index of refraction between the vitreous and the retina. Near the chorioretinal interface, a second peak in the intensity profile was observed corresponding to the photoreceptor inner-outer segment interface. The depth separation of these two peaks provided a measure of the total retinal thickness. Analysis of the images in our study confirmed the correspondence of the total retinal thickness profile with normal anatomy. Comparison of thickness measurements derived by the Matlab algorithm with values obtained by the standard OCT3 software indicated a high correlation. A bright band on the OCT images signified the inner retina and consisted of several retinal layers that were not separately resolved by the 10-m depth resolution of the commercial OCT3 instrument. Based on comparison of ultra high-depth resolution (1.4 m) OCT images with histology,11 the retinal layers that are likely included in 1060
AMERICAN JOURNAL
REFERENCES 1. Huang D, Swanson EA, Lin CP, et al. Optical coherence tomography. Science 1991;254:1178 –1181. OF
OPHTHALMOLOGY
JUNE 2005
2. Hee MR, Izatt JA, Swanson EA, et al. Optical coherence tomography of the human retina. Arch Ophthalmol 1995; 113:325–332. 3. Rogers AH, Martidis A, Greenberg PB, Puliafito CA. Optical coherence tomography findings following photodynamic therapy of choroidal neovascularization. Am J Ophthalmol 2002;134:566 –576. 4. Montero JA, Ruiz-Moreno JM, Tavolato M. Follow-up of age-related macular degeneration patients treated by photodynamic therapy with optical coherence tomography 3. Graefes Arch Clin Exp Ophthalmol 2003;241:797– 802. 5. Pires I, Bernardes RC, Lobo CL, Soares MA, Cunha-Vaz JG. Retinal thickness in eyes with mild nonproliferative retinopathy in patients with type 2 diabetes mellitus: comparison of measurements obtained by retinal thickness analysis and optical coherence tomography. Arch Ophthalmol 2002;120: 1301–1306. 6. Massin P, Erginay A, Haouchine B, Mehidi AB, Paques M, Gaudric A. Retinal thickness in healthy and diabetic subjects measured using optical coherence tomography mapping software. Eur J Ophthalmol 2002;12:102–108. 7. Goebel W, Kretzchmar-Gross T. Retinal thickness in diabetic retinopathy: a study using optical coherence tomography (OCT). Retina 2002;22:759 –767. 8. Drexler W, Morgner U, Ghanta RK, Kartner FX, Schuman JS, Fujimoto JG. Ultrahigh-resolution ophthalmic optical coherence tomography. Nat Med 2001;7:502–507. 9. Drexler W, Sattmann H, Hermann B, et al. Enhanced visualization of macular pathology with the use of ultrahigh-
VOL. 139, NO. 6
10.
11.
12.
13.
14.
15.
16.
QUANTITATIVE THICKNESS
OF
resolution optical coherence tomography. Arch Ophthalmol 2003;121:695–706. Gloesmann M, Hermann B, Schubert C, Sattmann H, Ahnelt PK, Drexler W. Histologic correlation of pig retina radial stratification with ultrahigh-resolution optical coherence tomography. Invest Ophthalmol Vis Sci 2003;44:1696 – 1703. Anger EM, Unterhuber A, Hermann B, et al. Ultrahigh resolution optical coherence tomography of the monkey fovea. Identification of retinal sublayers by correlation with semithin histology sections. Exp Eye Res 2004;78:1117– 1125. Giovannini A, Amato GP, Mariotti C, Scassellati-Sforzolini B. OCT imaging of choroidal neovascularisation and its role in the determination of patients’ eligibility for surgery. Br J Ophthalmol 1999;83:438 – 442. Kim SG, Lee SC, Seong YS, Kim SW, Kwon OW. Choroidal neovascularization characteristics and its size in optical coherence tomography. Yonsei Med J 2003;44:821– 827. Hee MR, Puliafito CA, Duker JS, et al. Topography of diabetic macular edema with optical coherence tomography. Ophthalmology 1998;105:360 –370. Hee MR, Baumal CR, Puliafito CA, et al. Optical coherence tomography of age-related macular degeneration and choroidal neovascularization. Ophthalmology 1996;103:1260 –1270. Schuman JS, Puliafito C, Fujimoto JG. Optical coherence tomography of ocular diseases, 2nd ed. Thorofare, NJ: Slack, 2004:21–27.
RETINAL LAYERS
1061
Biosketch Mahnaz Shahidi, PhD, is an associate professor of Physics in Ophthalmology at the University of Illinois at Chicago. Her area of research is development of optical imaging systems and techniques to provide for better visualization of retinal structures and quantitative assessment of abnormalities associated with eye diseases.
1061.e1
AMERICAN JOURNAL
OF
OPHTHALMOLOGY
JUNE 2005
● PURPOSE:
To report an image analysis algorithm that was developed to provide quantitative thickness measurement of retinal layers on optical coherence tomography (OCT) images. ● DESIGN: Prospective cross-sectional study. ● METHODS: Imaging was performed with an OCT3 commercial instrument in 10 visually normal healthy subjects. A dedicated software algorithm was developed to process the raw OCT images and detect the depth location of peaks from intensity profiles. Quantitative thickness measurements of three retinal layers, in addition to total retinal thickness, were derived. Total retinal thickness measurements obtained by the algorithm were compared with measurements provided by the standard OCT3 software. ● RESULTS: The total retinal thickness profile demonstrated foveal depression, corresponding to normal anatomy, with a thickness range of 160 to 291 m. Retinal thickness measured by the algorithm and by the standard OCT3 software were highly correlated (R ⴝ 0.98). The inner retinal thickness profile predictably demonstrated a minimum thickness at the fovea, ranging between 58 to 217 m along the 6-mm scan. The outer retinal thickness profile displayed a maximum thickness at the fovea, ranging between 66 to 107 m along the 6-mm scan. The photoreceptor outer segment thickness profile was relatively constant along the 6-mm scan through the fovea, ranging between 42 to 50 m. The intrasubject variabilities of the inner retina, outer retina, and photoreceptor outer segment thickness was 14, 10, and 6 m, respectively. ● CONCLUSIONS: Thickness measurements of retinal layers derived from OCT images have potential value for objectively documenting disease-related retinal thickness Accepted for publication Jan 6, 2005. From the Department of Ophthalmology and Visual Sciences, University of Illinois at Chicago, Chicago, Illinois. This study was supported by the Department of Veterans Affairs (Washington, DC); National Eye Institute EY14275 and EY01792 (Bethesda, MD); and an unrestricted fund from Research to Prevent Blindness (New York, NY). Inquiries to Mahnaz Shahidi, PHD, Department of Ophthalmology and Visual Sciences, University of Illinois at Chicago, 1855 West Taylor Street, Chicago IL 60612; fax: (312) 413-7366; e-mail: [email protected]
1056
©
2005 BY
abnormalities and monitoring progressive changes over time. (Am J Ophthalmol 2005;139:1056 –1061. © 2005 by Elsevier Inc. All rights reserved.)
I
MAGING AND QUANTITATIVE ASSESSMENT OF THE RET-
inal tissue is of value for the diagnosis of many retinal diseases. Optical coherence tomography (OCT) is an imaging technique that generates high-resolution crosssectional images of the retina and provides quantitative measurement of the total retinal thickness.1,2 OCT has been extensively applied for evaluating changes in retinal thickness resulting from various retinal diseases.3–7 Image analysis software that is provided by the commercially available OCT instrument allows measurement of total retinal thickness. Although intraretinal structures are visualized on OCT images, quantitative assessment of these structures is not obtainable by the instrument software. Recently, a new generation of OCT has been developed that has a higher depth resolution than the commercially available instrument8,9 and has been used to compare sublayers of the retina with histologic studies in the same eyes.10,11 This technology is in developmental stage, however, and not widely available because of its excessive cost. In this study, an image analysis algorithm was developed to measure the thickness of retinal layers from images obtained with the use of an OCT commercial instrument. Normal thickness measurements and intra- and intersubject variability were established from OCT images acquired in visually normal healthy subjects.
METHODS OCT IMAGING WAS PERFORMED USING AN OCT3 COMMER-
cial instrument (Carl-Zeiss Meditec, Dublin, California, USA). Six radial OCT scans, each 6 mm in length, were acquired. Each radial OCT scan was 1,024 pixels (2 mm) in depth and 512 pixels (6 mm) in length and thus had a depth resolution of 2 m/pixel and spatial resolution of 12 m/pixel. The vertical (scan 1 of 6) and horizontal (scan 4 of 6) OCT scans were analyzed. The raw gray scale OCT
ELSEVIER INC. ALL
RIGHTS RESERVED.
0002-9394/05/$30.00 doi:10.1016/j.ajo.2005.01.012
FIGURE 1. Outline of algorithm developed for analysis of optical coherence tomography (OCT) images. (A) OCT image in bitmap format obtained from a horizontal scan through the fovea obtained in one of the subjects in the study. (B) Exported OCT image in raw format. (C) Edge-enhanced processed image. (D) Location of the four peaks as identified on the edge-enhanced image: upper left, peak 1; upper right, peak 2; lower left, peak 3; lower right, peak 4. (E) Profiles of the identified peaks superimposed on the raw OCT image.
images were exported for analysis. A dedicated software program was developed in Matlab (Mathworks, Natick, Massachusetts, USA) for image analysis. The image analysis algorithm is outlined in Figure 1. A Gaussian function with normal distribution was generated and convolved with the raw image to derive an edge enhanced image. The image was cropped into a subimage of 15 (length) ⫻ 1,024 (depth) pixels. The subimage was averaged to derive an intensity profile. The profile was smoothed by median filtering and normalized. The points of maxima (peaks) and minima (valleys) of the profile were automatically identified. The first maximummost anterior peak of the intensity profile (peak 1) that had a value greater than 3 times the standard deviation of all peak intensities was identified, corresponding to the vitreoretinal interace. The second maximum peak, posterior to peak 1, was identified as peak 2, corresponding with the interface of inner and outer photoreceptor segments. The minimum valley between peaks 1 and 2, immediately anterior to peak 2, and with a value greater than three times the standard deviation of all valleys was identified as peak 3. The center between the 2 minimum valleys immediately posterior to peak 2 was identified as peak 4, corresponding to the chorioretinal interface. This procedure was repeated, each time shifting by 15 pixels (180 VOL. 139, NO. 6
m) along the image and generating another subimage. In this way, 34 coordinates (depth positions) for each of the four peaks were recorded along the 6-mm scan. Thickness profiles for three retinal bands were derived, in addition to the total retinal thickness, by calculating the peak separations and converting the pixel values to absolute thickness in microns. The separation between peak 1 (vitreoretinal interface) and peak 2 (inner and outer photoreceptor segments interface) was calculated along the 6-mm scan to provide a profile of the total retina (TR) thickness. The separation between peaks 2 and 3 provided a thickness profile of a retinal band designated as outer retina (OR) in this study. The separation between peaks 2 and 4 (chorioretinal interface) was calculated to provide a thickness profile of the photoreceptor outer segment (OS) layer. The separation between peaks 1 and 3 provided a thickness profile of the inner retina (IR). In each subject, OCT images from three vertical and three horizontal scans were analyzed, and the data were averaged to provide thickness profiles, along vertical and horizontal scans, respectively. Thickness profiles for each retinal layer, along the vertical and horizontal scans, were derived by averaging the profiles across all subjects.
QUANTITATIVE THICKNESS
OF
RETINAL LAYERS
1057
FIGURE 2. Orientations of the vertical and horizontal 6-mm OCT scans on a fundus image.
Intrasubject variability (reproducibility) was determined as the coefficient of variation (COV; standard deviation divided by the mean thickness) of three measurements in each subject, averaged over all subjects. Intersubject variability was determined as the standard deviation of thickness measurements across all subjects. OCT images were also analyzed using the standard OCT3 software. The TR thickness measurements obtained from three images in each eye were averaged. A TR thickness profile was generated by averaging the profiles in all subjects. The TR thickness profiles obtained by the Matlab algorithm and the standard OCT3 software were compared by linear regression analysis. Approval for use of human subjects was obtained from the Institutional Review Board of the University of Illinois at Chicago. The study was explained to the subjects, and informed consent was obtained according to the tenets of the Declaration of Helsinki. OCT imaging was performed in one eye of 10 visually normal subjects, 7 women and 3 men, five left and five right eyes. The subjects’ ages ranged from 25 to 56 years, with an average of 39 ⫾ 12 years (mean ⫾ SD).
FIGURE 3. Identification of retinal layers on an OCT image. The thickness of the total retina (TR), inner retina (IR), outer retina (OR), and photoreceptor outer segment (OS) are visualized on the OCT image. The outer retina (dark band) comprised outer nuclear cells and photoreceptor inner segments.
normal eye (Figure 2). The retinal layers as imaged by OCT are shown in Figure 3. The bands corresponding to IR, OR, OS, and TR are marked. The average TR thickness profiles derived from analysis of the vertical and horizontal scans in all subjects are shown in Figure 4. The average TR thickness profile demonstrated a foveal depression, corresponding to normal anatomy, with a thickness range of 160 to 291 m along the 6-mm scan. Mean retinal
RESULTS THE ORIENTATIONS OF THE VERTICAL AND HORIZONTAL
OCT scans are displayed on a fundus image of a visually 1058
AMERICAN JOURNAL
OF
OPHTHALMOLOGY
JUNE 2005
FIGURE 5. The relationship between total retinal thickness measured by the Matlab algorithm and by standard OCT3 software. Thickness measurements obtained by the two methods were highly correlated (R ⴝ 0.98).
FIGURE 4. Total retinal thickness profiles along the horizontal (top) and vertical (bottom) scans, averaged over all subjects and derived by the developed Matlab algorithm, corresponds to normal retinal anatomy. The standard deviation of measurements is indicated by error bars.
thickness was 243 ⫾ 35 m for the 6-mm vertical scans and 247 ⫾ 34 m for the horizontal scans. The mean intrasubject variability of total retinal thickness measurements derived by the COV was found to be 0.04, corresponding to a standard deviation of 10 m. The relationship between TR thickness profiles obtained by the Matlab algorithm and the standard OCT3 software is shown in Figure 5. The TR thickness profile mean obtained by Matlab algorithm and standard OCT3 software were 243 ⫾ 35 m and 232 ⫾ 35 m, respectively. The thickness measurements obtained by the two methods were highly correlated (R ⫽ 0.98). The thickness profiles of three retinal bands measured along the 6-mm OCT scan, averaged over all subjects, are shown in Figure 6. From the vertical scans, the mean thickness of profiles for the IR, OR, and OS retinal bands was 170 ⫾ 39 m, 78 ⫾ 10 m, and 47 ⫾ 2 m, respectively. From the horizontal scans, the mean thickness of profiles for the IR, OR, and OS retinal bands was 161 ⫾ 42 m, 81 ⫾ 13 m, and 46 ⫾ 2 m, respectively. The IR thickness profile demonstrated a minimum thickness at the fovea, ranging from 58 to 217 m along the 6-mm scan. The OR thickness profile displayed a maximum thickness of at the fovea, which ranged between 66 to 107 m along the 6-mm scan. The OS thickness profile was relatively constant along the 6-mm scan through the fovea, ranging between 42 to 50 m along the 6-mm scan. VOL. 139, NO. 6
FIGURE 6. Inner retina (IR), outer retina (OR), and photoreceptor outer segment (OS) thickness profiles along the horizontal (top) and vertical (bottom) scans, averaged over all subjects. The standard deviation of measurements is indicated by error bars.
The mean intrasubject thickness measurement variability (COV) of the IR, OR, and OS retinal bands on the vertical scans was found to be 0.08 (SD ⫽ 14 m), 0.13 (SD ⫽ 10 m), and 0.13 (SD ⫽ 6 m), respectively. The mean intrasubject thickness measurement variability (COV) of the IR, OR, and OS retinal bands on the
QUANTITATIVE THICKNESS
OF
RETINAL LAYERS
1059
horizontal scans was 0.09 (SD ⫽ 14 m), 0.15 (SD ⫽ 12 m), and 0.13 (SD ⫽ 6 m), respectively. The mean intersubject thickness measurement variability (COV) of the IR, OR, and OS retinal bands was found to be 0.11 (SD ⫽ 18 m), 0.14 (SD ⫽ 11 m), and 0.11 (SD ⫽ 5 m), respectively.
this band are ganglion cell axons, ganglion cell nuclei, inner plexiform, inner nuclear, and Henle fibers. Anatomically, the thickness of this band is expected to minimize at the foveal center because the thickness of most of these retinal layers becomes negligible at this point but maximizes within 1 mm nasal to the foveal center on the horizontal scan due to an increase in the ganglion cell axon and nerve fiber layer thicknesses in this region, consistent with the findings of our study. A dark band was observed, which signified the outer retina, as termed in this study. This retinal band comprised outer nuclear cells and photoreceptor inner segments. Our results indicate increased thickness of this band at the fovea as expected anatomically, partially due to the elongated foveal cone photoreceptors. A second dark band was identified bounded by chorioretinal interface and an intensity peak immediately anterior to the interface, which signified the region of photoreceptor outer segments.10 The intensity peak anterior to the chorioretinal interface on OCT images is likely to originate from the junction of the inner and outer photoreceptor segments.16 The thickness of this band was found to be relatively constant, consistent with an anatomically uniform thickness. In retinal diseases in which the photoreceptor cells have degenerated, a decrease in the thickness of outer retinal layers is anticipated. The reproducibility of outer retinal thickness measurements in normal subjects was found to be 10 m, indicating that thickness changes of 20 m can be detected over time with 95% confidence. The intersubject variability of the measurements was found to be 11 m, indicating that thickness abnormalities of 22 m can be detected with 95% confidence. In certain retinal degenerative diseases, the inner retinal thickness may also be altered. Based on the intra- and intersubject variability of the inner retinal thickness measurements, changes of 28 to 36 m can be detected with 95% confidence. The reproducibility of thickness measurements in patients with retinal pathologies or media opacities may be lower and needs to be established, however. The application of the described image analysis algorithm is limited to retinal pathologies that result in a relatively uniform thickness change in the retinal layers. Inhomogeneous retinal pathologies may cause local changes in the optical properties of the retinal layers. The influence of these changes on the peak detection algorithm and thickness measurements needs further investigation. Overall, thickness measurement of retinal layers has potential value for objectively documenting disease-related changes in retinal thickness and monitoring progressive changes over time.
DISCUSSION THE USEFULNESS OF OCT FOR IMAGING OF RETINAL PATHO-
logic changes in various diseases and detection of retinal thickness abnormalities has been established.6,7,9,12–15 The standard OCT3 software only offers measurements of total retinal thickness, however, and thus is limited in providing quantitative information on intraretinal structures. Development of new image analysis software that can measure the thickness of intraretinal structures provides additional information for documenting disease-related changes in the retina and has potential value for the diagnosis and monitoring of diseases in which depth-specific thickness abnormalities may be observed. In our study, development of an image analysis algorithm is reported that overcomes the limitation of the commercial OCT3 software and provides additional quantitative information that can be extracted from OCT images. The Matlab image analysis algorithm consists of processing of OCT images and detection of the depth location of peaks corresponding to high- and low-intensity bands. This algorithm can be applied to OCT macular images scanned at high resolution (1,024 ⫻ 512 pixels) by commercial OCT3 instrument. On the OCT images in normal eyes, three distinct retinal bands were identified. The thickness of these bands, in addition to the total retinal thickness, was measured. The intensity variation on OCT images stems from a combination of laser reflection, scatter, and absorption. At the vitreoretinal interface, a bright signal (peak in intensity profile) was detected because of a change in the index of refraction between the vitreous and the retina. Near the chorioretinal interface, a second peak in the intensity profile was observed corresponding to the photoreceptor inner-outer segment interface. The depth separation of these two peaks provided a measure of the total retinal thickness. Analysis of the images in our study confirmed the correspondence of the total retinal thickness profile with normal anatomy. Comparison of thickness measurements derived by the Matlab algorithm with values obtained by the standard OCT3 software indicated a high correlation. A bright band on the OCT images signified the inner retina and consisted of several retinal layers that were not separately resolved by the 10-m depth resolution of the commercial OCT3 instrument. Based on comparison of ultra high-depth resolution (1.4 m) OCT images with histology,11 the retinal layers that are likely included in 1060
AMERICAN JOURNAL
REFERENCES 1. Huang D, Swanson EA, Lin CP, et al. Optical coherence tomography. Science 1991;254:1178 –1181. OF
OPHTHALMOLOGY
JUNE 2005
2. Hee MR, Izatt JA, Swanson EA, et al. Optical coherence tomography of the human retina. Arch Ophthalmol 1995; 113:325–332. 3. Rogers AH, Martidis A, Greenberg PB, Puliafito CA. Optical coherence tomography findings following photodynamic therapy of choroidal neovascularization. Am J Ophthalmol 2002;134:566 –576. 4. Montero JA, Ruiz-Moreno JM, Tavolato M. Follow-up of age-related macular degeneration patients treated by photodynamic therapy with optical coherence tomography 3. Graefes Arch Clin Exp Ophthalmol 2003;241:797– 802. 5. Pires I, Bernardes RC, Lobo CL, Soares MA, Cunha-Vaz JG. Retinal thickness in eyes with mild nonproliferative retinopathy in patients with type 2 diabetes mellitus: comparison of measurements obtained by retinal thickness analysis and optical coherence tomography. Arch Ophthalmol 2002;120: 1301–1306. 6. Massin P, Erginay A, Haouchine B, Mehidi AB, Paques M, Gaudric A. Retinal thickness in healthy and diabetic subjects measured using optical coherence tomography mapping software. Eur J Ophthalmol 2002;12:102–108. 7. Goebel W, Kretzchmar-Gross T. Retinal thickness in diabetic retinopathy: a study using optical coherence tomography (OCT). Retina 2002;22:759 –767. 8. Drexler W, Morgner U, Ghanta RK, Kartner FX, Schuman JS, Fujimoto JG. Ultrahigh-resolution ophthalmic optical coherence tomography. Nat Med 2001;7:502–507. 9. Drexler W, Sattmann H, Hermann B, et al. Enhanced visualization of macular pathology with the use of ultrahigh-
VOL. 139, NO. 6
10.
11.
12.
13.
14.
15.
16.
QUANTITATIVE THICKNESS
OF
resolution optical coherence tomography. Arch Ophthalmol 2003;121:695–706. Gloesmann M, Hermann B, Schubert C, Sattmann H, Ahnelt PK, Drexler W. Histologic correlation of pig retina radial stratification with ultrahigh-resolution optical coherence tomography. Invest Ophthalmol Vis Sci 2003;44:1696 – 1703. Anger EM, Unterhuber A, Hermann B, et al. Ultrahigh resolution optical coherence tomography of the monkey fovea. Identification of retinal sublayers by correlation with semithin histology sections. Exp Eye Res 2004;78:1117– 1125. Giovannini A, Amato GP, Mariotti C, Scassellati-Sforzolini B. OCT imaging of choroidal neovascularisation and its role in the determination of patients’ eligibility for surgery. Br J Ophthalmol 1999;83:438 – 442. Kim SG, Lee SC, Seong YS, Kim SW, Kwon OW. Choroidal neovascularization characteristics and its size in optical coherence tomography. Yonsei Med J 2003;44:821– 827. Hee MR, Puliafito CA, Duker JS, et al. Topography of diabetic macular edema with optical coherence tomography. Ophthalmology 1998;105:360 –370. Hee MR, Baumal CR, Puliafito CA, et al. Optical coherence tomography of age-related macular degeneration and choroidal neovascularization. Ophthalmology 1996;103:1260 –1270. Schuman JS, Puliafito C, Fujimoto JG. Optical coherence tomography of ocular diseases, 2nd ed. Thorofare, NJ: Slack, 2004:21–27.
RETINAL LAYERS
1061
Biosketch Mahnaz Shahidi, PhD, is an associate professor of Physics in Ophthalmology at the University of Illinois at Chicago. Her area of research is development of optical imaging systems and techniques to provide for better visualization of retinal structures and quantitative assessment of abnormalities associated with eye diseases.
1061.e1
AMERICAN JOURNAL
OF
OPHTHALMOLOGY
JUNE 2005