High-resolution axial measurements of ocular tissues

Transcription

1 The Interpretation of Optical Coherence Tomography Images of the Retina Devinder Singh Chauhan and John Marshall PURPOSE. To determine the relationship between optical coherence tomography (OCT) images of the retina and retinal substructure in vitro and in vivo. METHODS. In vitro, OCT images of human and bovine retina were acquired after sequential excimer laser ablation of the inner retinal layers. Measurements of bands in the OCT images were compared with measurements of retinal layers on histology of the ablated specimens. In vivo, OCT images were acquired of retinal lesions in which there was a displacement of pigmented retinal pigment epithelial (RPE) cells: retinitis pigmentosa and laser photocoagulation (eight eyes each). RESULTS. The mean thickness of human inner OCT bands (131 m; 95% confidence interval [CI], m) was 7.3 times that of the retinal nerve fiber layer (RNFL). This band persisted despite ablation greater than 140 m. The inner aspect of the outer OCT band corresponded to the apical RPE, but the mean thickness of this band in human tissue (55 m; 95% CI, m) was 2.6 times the thickness of the RPE choriocapillaris complex. OCT measurement of total retinal thickness was accurate (coefficient of variance, 0.05) and precise (coefficient of correlation with light microscopy, 0.98). Hyperpigmented lesions gave rise to high signal, attenuating deeper signal; hypopigmented lesions had the opposite effect on deeper signal. CONCLUSIONS. The inner band is not RNFL specific, partly consisting of a surface-related signal. The location, not thickness, of the outer band corresponds to RPE melanin. Given the additional effect of polarization settings, precise OCT measurement of specific retinal layers is currently precluded. (Invest Ophthalmol Vis Sci. 1999;40: ) High-resolution axial measurements of ocular tissues have become important in both the medical and surgical management of disease. For example, the advent of the intraocular lens led to a demand for precise measurement of the axial length of the globe and encouraged the development of ultrasonic scanning systems. 1 More recently, an increasing requirement for the measurement of retinal thickness has become apparent, particularly in relation to the management of macular disease. 2 To date, estimations of retinal thickness have been achieved by stereoscopic biomicroscopy, 3 and measurements have been obtained using lasergenerated slits 4 or scanning slit devices. 5 The advent of optical coherence tomography (OCT) offers the theoretical possibility of high-resolution measurements of both retinal thickness and the dimensions of the retinal component layers. Detailed accounts of the underlying principles of OCT have been reported elsewhere. 6 8 In summary, however, images are generated as a result of the interaction between a From the Department of Ophthalmology, Rayne Institute, United Medical and Dental Schools of Guy s and St. Thomas, St. Thomas Hospital, London, United Kingdom. Supported by a grant from the Lady Anne Allerton Fund and forms part of a contribution toward a thesis for a Doctor of Medicine degree (DSC) at the University of London. Submitted for publication October 16, 1998; revised March 3, 1999; accepted April 12, Proprietary interest category: N. Corresponding author: Devinder Singh Chauhan, Department of Ophthalmology, Rayne Institute, United Medical and Dental Schools of Guy s and St. Thomas, St. Thomas Hospital, Lambeth Palace Road, London SE1 7EH, UK. partially coherent beam of optical radiation and tissue components. The most important optical phenomena in OCT signal generation are the scatter, 9 reflection, 10 and absorption of incident light. In current biomedical applications, near infrared radiation is used almost exclusively as a source of illumination. The resolution of spatial information within OCT images has both theoretical and practical limitations. In axial, or z-plane, measurements the resolution is 17 m in air, 6 which is inversely related to the coherence length of the superluminescent diode light source. In biologic tissue it is somewhat larger but always better than the 120 m currently available with ultrasound. 11 In the x y plane, the maximum resolution is diffraction limited by the properties of the ocular media and is approximately 25 m. Empirically, however, the overall resolution across the image is limited by the design of the instrument used (Humphrey Instruments, San Leandro, CA), in which 100 individual z-axis scans are required to generate each image, irrespective of the field of the image. Commercial instruments became available late in 1996, and several groups have published pseudocolored images displaying several layers. These layers have stereospatial locations that apparently correlate with the different cell layers of the retina and choroid. Therefore, the spatial separation of the colored bands in these pseudocolor images is similar to that of a histologic section. There are four pseudocolored bands in images of extrafoveal retina that, on passing from the vitreous surface toward the sclera, have the following sequence: redwhite, yellow-green, black, and red-white. Previous reports have assumed a correlation between the four bands and the four cell layers of the retina seen on histology. 6,7,12 However, the dimensions of the pseudocolor bands do not display the Investigative Ophthalmology & Visual Science, September 1999, Vol. 40, No Copyright Association for Research in Vision and Ophthalmology

2 IOVS, September 1999, Vol. 40, No. 10 Interpreting Retinal Optical Coherence Tomography Images 2333 same ratios as the cell layers in histologic sections. It is therefore extremely difficult to assign the pseudocolor bands in the OCT images to specific anatomic components. In previous reports the innermost (red-white) band was thought to correlate with the retinal nerve fiber layer (RNFL), whereas the outermost (red-white) band was thought to be some function of the retinal pigment epithelium choriocapillaris complex (RPE-CC) 6,7,12 (Fig. 1A). Crude measurements of the separation between the innermost borders of these bands approximate to the average thickness of the neural retina. However, given the pixelated nature of the image, it is difficult to determine precisely the location of the borders of these and the other pseudocolored bands. Given a theoretical axial resolution of 14 m, discrimination between tissue layers in areas where anatomic features are well defined might be expected. In the inner retina, however, there appears to be little significant change in the innermost band at the macula where the RNFL is reduced and the ganglion cell layer increases from one to three or more nuclei in thickness. 8,12,13 Discrimination in the pseudocolor images between the pigment-rich RPE and the underlying blood vessels also appears difficult. In view of the number of recently published papers using OCT to measure changes in retinal dimensions as a function of disease, 7,8 it is surprising that only one empiric attempt has been made to relate pseudocolor banding to structural components. 14 In this study the histologic images were manipulated in an attempt to superimpose one on the other, using the inner limiting membrane as an inner reference location. The outer reference chosen was the RPE. An assumption was made that this layer correlated with the superficial border of the outer band. A more direct correlation could be obtained in two ways: either by empirically changing retinal dimensions in a controlled fashion or by studying pathologic conditions in which a highly visible marker cell is known to have changed its location within the retina. The former could be performed using the excimer laser, which has the ability to remove large areas of tissue with precise control of ablation depth. If the ablation rate of retina is similar to that of cornea, then at a radiant exposure of 180 mj/cm 2 each laser pulse would remove approximately 0.25 m of tissue. 15 The latter experimental approach could be investigated by examining conditions in which RPE cells migrate into known regions 16,17 of the neural retina. The present article describes an experimental investigation of OCT images in which both approaches have been taken. FIGURE 1. (A) OCT parameters measured. Normal retina. Schematic diagram of an OCT scan of retina with the major high-signal OCT bands shown. The parameters measured were the thicknesses of the inner (a) and the outer (c) bands of high signal, the total retinal thickness, (b) and the distance between the outer margins of the inner and outer bands (d). Laser lesion: (e) and (f) represent the extent of the pale rim and pigmented center of a retinal laser lesion, respectively. The parameters (b ) and (c ) at the pigmented center of a laser lesion correspond to (b) and (c) in normal retina. (B) OCT thickness measurements of human retinal laser lesions. The graph shows data collected from 20 lesions. The location of the boundaries of both the high-signal outer band (c ) and total retinal thickness (b ) are shown. Measurements of thickness and location are relative to the location of the innermost aspect of the outer band in surrounding normal tissue, seen to the right of the parallel dashed lines. Error bars, 95% confidence limits. METHODS In the first part of this study we observed changes in OCT images of cadaveric human and bovine retina in vitro after dissection and controlled removal of superficial retinal layers. A pilot study using human cadaveric retinal tissue had already shown the lamination pattern and dimensions of OCT images of isolated retina to be the same as human retina in vivo (Chauhan DS, Marshall J, unpublished data). For experimental studies, bovine tissue was chosen because it was readily available and had a number of features of its vitreoretinal architecture similar to those of human retina. In addition, bovine retina has tapetal and nontapetal regions that differ in thickness and histologic structure, tapetal retina being thicker. The tapetum separates the CC from the medium and large vessels of the choroid, and there is no melanin pigment within the RPE of this region. 18 All studies were performed in accordance with the ARVO Statement for the Use of Animals in Ophthalmic and Vision Research and the Declaration of Helsinki. The second part of the study was based on the observation that, for light passing through the retina, the site of greatest change in optical properties is the layer of melanin within the RPE. We therefore examined lesions in which abnormal changes in the distribution of pigmented cells were apparent. We chose retinitis pigmentosa (RP) as a condition in which pigmented RPE cells migrate forward into the inner layers of the neural retina and panretinal photocoagulation as a model for the redistribution of RPE pigmentation within the plane of the RPE. These studies were in accordance with the Declaration of Helsinki recommendations. OCT Imaging. Images were acquired using an optical coherence tomography scanner (Humphrey Instruments), which directed a beam of light in a horizontal plane through the pupil. Each composite OCT image or scan consisted of a linear array of 100 juxtaposed individual z-axis scans, each with a depth of 2 mm. The thickness of each individual z-axis scan was 13 m, and the separation between them increased with the field of the image. The length of the OCT scan line in the x y plane could be varied, ranging from 2.5 to 19 mm in air at a working distance of 43 mm. This represented a scanning angle of between 3 and 25. For a flat specimen, this results in an overestimate of thickness by a factor of 1.00 to 1.024, respectively, at the lateral extremes of a scan. For this reason, the shortest scans possible were used in the in vitro studies, even though posterior pole specimens maintained the concavity of the eye. For an emmetropic eye the system-corrected range for scan length

3 2334 Chauhan and Marshall IOVS, September 1999, Vol. 40, No. 10 in the x y plane of the retina was 1.13 to 9.08 mm. The angular orientation of the scan line could also be controlled within the x y plane of the tissue examined. The device had an internal fixation light visible to the patient but not on the operator s viewing screen. A second light, generated by a HeNe laser, was seen on the viewing screen and could be moved under the control of the operator to fall on any given retinal feature. The spatial separation between this light and the ends of the scan line was stored by the computer for each scan orientation. Thus, on repeat visits, if the operator relocated the HeNe light over the original retinal feature, subsequent scans would always be at the same location and orientation. The acquisition time for each composite OCT image was 0.9 seconds, independent of scan length and orientation. The images produced were dependent on the optical properties of tissues in the z-plane and underwent processing by the commercial software into a logarithmic pseudocolor scale in which white, red, yellow, green, blue, and black represented the range of signal intensity from high to low. All scans obtained with this instrument were accompanied by a digital image of the tissue recorded with the scan line superimposed. The power incident on the eye was constant throughout (750 W). Imaging began with crude focusing of the image on the operator s viewing screen. After this, the continuously scanning real-time OCT image was centered on the computer screen and the focusing knob adjusted by small increments until the image was at its brightest throughout the tissue. The polarization setting was then varied for each image set to optimize the signal intensity for the inner band, because the outer band varied little with polarization. This was done to maximize the contrast in signal intensity of the inner part of the image and therefore the clarity of borders. Finally, the focusing knob was readjusted, if necessary, according to the criteria described. In vivo examinations were performed after pupillary dilation with 1% tropicamide and 10% phenylephrine. To avoid degradation of images, only scans with minimal eye movement were recorded for later analysis. Internal fixation was used as far as possible, but external fixation was occasionally necessary to scan peripheral lesions. 19 Subjects were asked to blink between acquired images to minimize corneal drying, which may have caused both discomfort for the subject and a degradation of the OCT image. Because the measurements of interest in this study were all in the z-plane and taken from several individual z-axis scans in each OCT image, z-axis alignment software was not used. In Vitro Studies OCT Imaging. For the purposes of in vitro examination of tissues, the OCT scanner was modified by placing a plane mirror above and at 45 to the objective lens, thus allowing the acquisition of images of horizontally oriented tissues. Extensive validation studies demonstrated that the images acquired with this modification were identical in pattern and dimensions to those without it. When comparing the measurement of the thickness of glass slides and signals from glass air interfaces with and without the plane mirror, the coefficients of variation of measurements were 0.04 and 0.03, respectively. The correlation between measurements made with and without the mirror was very strong (R ). To avoid the inherent inversion of vertical images obtained from such an optical system, scans were taken in the transverse horizontal axis. Specimens. Fresh bovine eyes were obtained from an abattoir, in accordance with European Union statute 716/ Four cadaveric human eyes were obtained from the National Transplant Service (Bristol, UK) with a mean postmortem time of 50 hours. Retinal tissue was examined from two further human eyes immediately after their enucleation for choroidal malignant melanoma. The latter and the bovine eyes were intact, whereas the cadaveric eyes from the eye bank were supplied after corneal excision. Globes were hemisected by a circumferential penetrating incision 5 mm posterior to the sulcus sclerae. With the anterior globe facing downward, the posterior globe was then separated from this and the vitreous by gently lifting and rotating the optic nerve. 21 In bovine eyes full-thickness specimens were obtained from both tapetal and nontapetal regions using an 8-mm trephine. Similarly, specimens were trephined from human eyes to include attached retina from the posterior pole in cadaveric eyes and from the quadrant opposite the choroidal melanoma in the tumor-affected eyes. OCT Image Acquisition. For OCT imaging, these fullthickness specimens were placed in a petri dish and covered to a depth of 5 mm with fresh normal saline. Using the modified scanner, multiple contiguous scans (1.7 mm) were performed across the entire diameter of each sample, using the HeNe relocation feature to position scans end to end. All scans on each specimen were performed at the same focal plane using the same power and polarization settings and repeated three times. All scans were acquired by a single observer in these studies except when stated otherwise. Specimens were then fixed in 2.5% glutaraldehyde buffered in 0.1 M sodium cacodylate containing 1 mg/ml calcium chloride with a final ph of 7.4. In this solution, the areas scanned before fixation were then rescanned at the same focal plane and using the same settings at 1, 15, 35, and 60 minutes and 24 hours after fixation. To determine the dependence of measurements derived from OCT images on the system settings for polarization, the results of 10 independent observers were obtained. Using a single fixed specimen of human extramacular retina, each of the observers was asked to obtain the best image by altering the polarization setting. All other control settings remained constant. Observers were masked to each other, and the whole image sequence was performed within a 1-hour period. Only two of these observers had any previous experience in OCT image acquisition. Excimer Laser Ablation. Each full-thickness trephined specimen was removed from the glutaraldehyde solution and placed on a fixative-soaked piece of filter paper in a covered petri dish. The petri dish was then placed on a laboratory jack under the exit port of an excimer laser (Apex Plus, Summit Technology, Boston, MA). The petri dish was manipulated so that the specimen was centered using the HeNe laser-aiming beams. After the petri dish was uncovered, surface fluid was removed using the capillary action of a Wechsel sponge, and the HeNe beams were then focused on the surface of the tissue. With a radiant exposure of 180 mj/cm 2 per pulse and the program designed for phototherapeutic keratectomy, concentric ablations were performed. The initial ablation zone diameter used was 6.5 mm, and this was reduced by 0.5 mm for each successive ablation to a minimum diameter of 1 mm. This resulted in 11 steps between zones. Fifty pulses were applied per ablation zone for both the

4 IOVS, September 1999, Vol. 40, No. 10 Interpreting Retinal Optical Coherence Tomography Images 2335 bovine and human specimens. A typical procedure took between 12 and 20 minutes and was limited by the reprogramming time for each successive ablation zone. The lid of the petri dish was removed only during the periods of excimer ablation in an attempt to minimize drying of the tissues. Immediately after the completion of the ablation process, the specimens were reimmersed in glutaraldehyde solution. The contoured sample was then processed for light and scanning electron microscopy. Measurements on Microscopy. Retinal features were measured by two independent observers using a light microscope with a calibrated measuring eyepiece. In all specimens the baseline thicknesses of individual layers of choroid and retina were measured in an area that had not been ablated. In addition, the total remaining thickness of retina in each ablation zone was measured, as was the height of each step. OCT Image Analysis. The pseudocolored images produced were presented as raw images in the first instance and were then postprocessed using the manufacturer s programs. The raw images and the normalized, median-smoothed, and gaussian-smoothed images were all converted to tagged information file format (TIFF) using a screen-capture program (Jas- Capture Version Shareware 2.0; JASC, Eden Prairie, MN). These files were then imported into an image-processing program (Photoshop, ver. 4.0, Adobe, San Jose, CA) and magnified without altering the image pixel dimensions. The borders of the inner and outer bands of high signal were defined as the point at which the signal s false color changed from white or red to a color indicating lower signal intensity. To determine the variation in measurements between the eight independent observers, four image parameters were measured on the unablated specimens. These were the distance between the inner borders of the inner and outer bands, the overall thickness of the inner band, the overall thickness of the outer band and, finally, the distance between the outer borders of the inner and outer bands (Fig. 1A). All measurements were repeated eight times by a single experienced image analyzer for each scan. For the ablated specimens, measurements of the thickness of the colored bands were made using the measuring facility of the image-processing program. The exact location of the borders was determined by the operator, as described. The overall thickness of the unablated retina and the overall thickness at each ablation zone were also measured. The former was taken to be the distance between the superficial borders of the inner and outer bands in nonablated tissue. 12,14 These measurements were converted to micrometers using a manufacturer s conversion factor of 4 m per z-axis pixel (Humphrey Instruments). The thickness of the remaining retina in each ablation zone was expressed in relation to the thickness of the unablated retina by subtracting the sum of the heights of the steps peripheral to the zone in question. The height of each step was measured as the distance between the extrapolated superficial borders of the inner bands. The same measurements were also attempted using the pixelated calipers available on the OCT scanner s analysis software. However, such measurements were abandoned because of the technical limitations of the system. In Vivo Studies Subjects. The 10 independent observers each imaged the same area of extramacular retina of a single normal subject. They again altered the polarization setting only to obtain the best image. For a single-observer study, 10 subjects with normal retinas were then selected. Each was less than 30 years of age, had uncorrected visual acuity of 6/6 or better, and had no significant medical or ophthalmic history. In this group a series of radial 3-mm linear scans was obtained, each centered on the fovea. For each of these subjects, these scans were optimized to provide the maximal signal intensity to the inner band using all available control settings. After these images were acquired for each subject, a further scan was undertaken with the polarization adjusted to provide the minimum inner band signal intensity. All other control settings were maintained in a constant position. Four patients with RP were examined, two with Usher s syndrome (US; 42 and 73 years old) and two with autosomal dominant (AD) RP (47 and 62 years old). Four patients with panretinal laser photocoagulation for vascular disease were also examined. All had lesions of at least 3 months duration with a hyperpigmented center and pale rim. All lesions were produced with an argon laser (Novus 2000, Coherent, Palo Alto, CA) with a spot size of 200 m, pulse duration of 0.1 second, and power of 170 to 300 mw. OCT Image Acquisition. Images were optimized for each scan to produce the highest intensity and definition to the inner and outer bands by manipulating the polarization control. Each control subject had eight 3-mm scans taken, centered on the foveola and at 0, 30, 45, 60, 90, 120, 135, and 150. The patients with RP had scans taken in two configurations. The first was a short ( mm) linear scan taken through the center of one or more bone spicule lesions with normally pigmented intervening retina. The second was a longer linear scan spanning across the macula and the surrounding area of hypopigmentation when present. Photocoagulation lesions had short linear scans performed passing through their centers. In each case the scans were extended to include adjacent retina of normal appearance. OCT Image Analysis. In the study, to derive the interobserver error, the same parameters were measured on the in vivo images as for the in vitro specimens. In the single-observer study of 10 normal subjects, these parameters were measured both in the center of the foveola and the edge of the foveal pit for the images with the polarization maximized and minimized. There were two aspects to the analysis of OCT images of 54 areas of bone spicule pigmentation. First, any discrete areas of higher signal within the inner band were identified and their locations in the x y plane related to those of bone spicule pigmentation. The depth of each lesion was measured on OCT as the distance between the superficial border of the inner band and the center of the discrete high signal. Second, discontinuities in the outer band were identified and measured in the x y plane of the image. The location of each discontinuity was also related to that of the bone spicule pigmentation. In the scans passing through the macula and surrounding hypopigmented retina, the thickness of the whole retina and outer band in each region was also measured. For the OCT scans of laser lesions, a number of parameters were measured in both the vertical (z) and the horizontal (x y) axes. The z-axis measurements were made using the same measuring facility described earlier. They were undertaken at the center of the lesions and on adjacent tissue that appeared

5 2336 Chauhan and Marshall IOVS, September 1999, Vol. 40, No. 10 TABLE 1. Statistical Parameters for the Measurements of Inner and Outer High-Signal Band Thicknesses and Total Retinal Thickness in vitro and in vivo. Inner Band Thickness ( m except italics) Outer Band Thickness ( m except italics) Total Retinal Thickness ( m except italics) 10 Observers: Single specimen (in vitro) All readings Mean Standard deviation % Confidence interval Coefficient of variation Between observers Mean Standard deviation % Confidence interval Coefficient of variation Single observer: 10 subjects (in vivo) Foveola: minimum polarization Mean Standard deviation % Confidence interval Coefficient of variation Foveola: maximum polarization Mean 23* Standard deviation % Confidence interval Coefficient of variation Clivus: minimum polarization Mean Standard deviation % Confidence interval Coefficient of variation Clivus: maximum polarization Mean 128* Standard deviation % Confidence interval Coefficient of variation All figures, except those in italics, are in micrometers. The thickness of individual bands varied by up to 30% between individual measurements and 18% among observers in the in vitro study. The in vivo study showed that altering polarization has a profound effect on the measured thickness of the inner high-signal band in particular. * P normal on both sides of the lesion. The depression of the center of the lesion relative to the surrounding retina was also measured. The x y plane measurements were made after vertical borders had been superimposed on the images (Fig. 1B). These borders were defined by the sharp x y-axis transition in the signal intensity deep to the outer band. Pixel values in the x y plane were converted to micrometers by correcting for the given length of scan line, which was taken as that on the OCT display. Because the maximum refractive error of the subjects was 2.5 D of myopia, any discrepancy between real and actual scan length was small. RESULTS In Vitro Studies The bovine eyes were well preserved with the retina attached. The cadaveric human eyes often had retinas that had partially detached after death. Small serous detachments were also noted in the freshly enucleated human eyes adjacent to the melanoma. Glutaraldehyde fixation caused no significant change in either the pattern or measured thickness of the individual OCT bands or the total retinal thickness. The signal intensity of each band increased, however, and reached a maximum within 35 minutes of fixation. Signal intensity and dimensions in OCT images were found to vary with control settings. Using a constant power and an optimally focused image, manipulation of the polarization control resulted predominantly in variations in the inner band. The setting of the polarization control varied among individual observers in their attempts to obtain an optimized image. Statistical parameters of the variation among observers are shown in Table 1, in which altering polarization settings can be seen to result in a coefficient of variation in measurements of up to The thicknesses of bovine retina varied between eyes and regions of retina. Typically, tapetal retina was thicker than nontapetal and had a relatively thicker RNFL. In a representative eye, the mean thickness of tapetal retina was 235 m (95% CI, m) measured using OCT and 248 m (95%

6 IOVS, September 1999, Vol. 40, No. 10 Interpreting Retinal Optical Coherence Tomography Images 2337 CI, m) using light microscopy. The corresponding values for nontapetal retina were 120 m (95% CI, m) and 128 m (95% CI, m), respectively. The relationship between OCT and light microscopic measurements of total retinal thickness over all specimens had a strong linear correlation (r ) with a gradient of 1.05 and an intercept on the y-axis (OCT measurement) of 6.9 m. In all human and bovine eyes, the location and thickness of the inner band of high OCT signal corresponded, on light microscopy, to the sum of RNFL, ganglion cell layer (GCL), inner plexiform layer (IPL), and part of the inner nuclear layer (INL). The mean thickness of the inner band was 7.3 times that of the RNFL (Fig. 2). In 25% of the eyes, the outer plexiform layer (OPL) was included within this band. Similarly, the outer band of high signal corresponded to the RPE, CC, and tapetum in tapetal retina. However, in bovine nontapetal and human retina this band corresponded to the RPE, CC, and just under half of the choroid. Here, the mean thickness of the outer band was 2.6 times that thickness of the RPE-CC complex. The thickness of the dark band of low OCT signal, between the inner and outer bands, was greater than that of the inner and outer segments of the photoreceptors in tapetal retina but less in bovine nontapetal and human retina. Excimer-Ablated Specimens. During the process of ablation, no changes were observed in transparency of the retina. The steps were always clearly seen with the microscope of the excimer laser. The mean depth of ablation per pulse in glutaraldehyde-fixed retina was 0.49 m (95% CI, m), measured by light microscopy (Fig. 2). There was no significant difference in ablation rate between human and bovine retina. OCT Signal Changes. With the progressive ablation of tissue, changes in the location and thicknesses of one or both of the major OCT bands were observed in all specimens. Inner Band of High OCT Signal. The number of pulses required to remove a depth of retina corresponding to the inner band measured in unablated tissue varied from one specimen to another but was typically between 150 and 250. This corresponded to between 74 and 123 m. In practice, it was not possible to remove the inner band by ablation, because it migrated down through the tissue with progressive laser pulses. Although the band was always present, its width decreased linearly with ablation until the OPL had been reached. After this point the band remained roughly constant with a mean thickness of 32 m (95% CI, m) but moved deeper, related to the tissue surface, with subsequent ablation pulses (Fig. 2). Outer Band of High OCT Signal. The outer band maintained its location and thickness after ablation of retina in all specimens except bovine tapetal retina. In the latter, the thickness of this band increased with progressive ablation of overlying retinal tissue. In both human and bovine specimens the inner border of the outer band remained within 28 m ofthe location of the apical RPE, as defined by light microscopy (Fig. 2). Dark Band of Low OCT Signal. The thickness of this band was stable in all specimens until the ablation process reached the level corresponding to the ONL on light microscopy (Fig. 2). After this, a linear reduction in the thickness of this band was observed. In Vivo Studies The reproducibility of OCT measurements of total retinal thickness among the 10 observers was high (coefficient of variance, 0.05). It was lower (coefficient of variance, 0.18) for the distance between the outer borders of the major bands (Fig. 1A). In most subjects the inner band was absent at the foveola and occupied half the total retinal thickness at the edge of the foveal pit (Table 1). The mean total retinal thickness at the foveola was 159 m (95% CI, m), and that at the edge of the foveal pit was 258 m (95% CI, m). The respective outer band thicknesses were 81 m (95% CI, m) and 71 m (95% CI, m) and differed significantly (P 0.05). At the extreme settings of polarization, the overall retinal thickness and the thickness of the outer band remained constant. By contrast, there was a highly significant change in the thickness of the inner band, which varied from a mean thickness of 11 m (95% CI, 1 23 m) at the minimum setting to 128 m (95% CI, m) at the maximum. The reproducibility of these measurements was very low, with a coefficient of variance of RP. In scans of patients with RP, although inner and outer bands could be readily identified, the inner band was less intense. Fifty-four areas of bone spicule hyperpigmentation were imaged in the four patients. All scans passing through these lesions had discrete regions of very high signal coincident with the bone spicules, and deep to the innermost border of the inner band (Fig. 3A). There was also a discrete reduction in signal intensity of the outer band under such high-signal areas. Three of the patients had high-signal regions imaged with a median depth of 48 m, and the other patient had similar regions at a median depth of 116 m. The patient with the deeper signals was 47 years old with AD RP. Two patients (one with US and one with AD RP) had a central macular area of normal appearance with a surrounding area of hypopigmented retina. Scans spanning these areas had three distinctive features coincident with the hypopigmented retina (Fig. 3B). First, the inner band of the OCT image of the hypopigmented retina had an increased signal. Second, there was a reduced distance between the superficial borders of the inner and outer bands (i.e., total retinal thickness). Finally, the outer band was much thicker than that of the central macula. Laser Lesions. The mean thickness of the normal retina in scans passing through laser lesions was 170 m (95% CI, m). The regions of OCT images corresponding to the hyperpigmented centers of laser lesions were thinner, with a mean thickness of 138 m (95% CI, m), and their surface was depressed by a mean of 19 m (Figs. 1B, 3C). There was also a significant increase in the mean thickness of the outer band (64 m) compared with the normal surrounding retina (41 m; P 0.05). The pale rim of the laser lesions corresponded to an increase in the signal intensity from deeper tissues. DISCUSSION The optical coherence tomographic imaging of tissues is dependent on their optical properties. In any illuminated tissue the only structures to generate high signal are those that return light close to or along the axis of the incident light. Tissue

7 2338 Chauhan and Marshall IOVS, September 1999, Vol. 40, No. 10 FIGURE 2. Excimer laser ablation of retinal tissue. (A) Scanning electron micrograph of the surface of a hemisected full-thickness trephine of the posterior pole of a bovine eye after stepwise ablation using an excimer laser. (B) Light micrograph of the same specimen. The black horizontal bar marker represents 100 m. (C) OCT scan of the same specimen as (A) and (B) above. With progressive ablation, from left to right on this scan, the inner band becomes thinner initially and then persists and migrates downward. Finally, it becomes indistinguishable from the outer band, which maintains its location. (D) Measurements taken from OCT scans and histology of a single representative human retinal specimen. The thick, solid, horizontally oriented lines represent the location of the innermost borders of the inner and outer high-signal OCT bands. The thick vertical lines represent the thickness and location of the inner and outer high-signal OCT bands. The narrow horizontal lines mark the borders of the individual layers of the retina measured on histologic examination of the specimen scanned. P, photoreceptor inner and outer segments, CC, choriocapillaris, CH, choroid. Error bars, 95% confidence limits. Magnification, 180. components that absorb light or direct it away from the OCT detector have correspondingly low signal intensities. Thus, the intensity of a recorded signal is related to the proportion of low-coherence light returned (reflectance) to the detector. The optical phenomena involved in the generation of OCT signals in tissues are likely to be reflection, scatter, and birefringence. 9,10 Theoretical and empiric studies on multilayered model systems have demonstrated that the change in refractive index at an interface is responsible for most of the signal, the next most important factor being scatter. 22 In these model studies, refractive index changes gave rise to specular reflection and sharp signal peaks. Scattering properties contributed to the band thickness. It was found that the more a medium scattered light, the more discrete were its deep borders, and

8 IOVS, September 1999, Vol. 40, No. 10 Interpreting Retinal Optical Coherence Tomography Images 2339 FIGURE 3. OCT scans of retinal lesions with altered melanin location. (A) Bone spicule in RP. The OCT scan on the right corresponds to the location of the white line (x y), on the photograph on the left. This passes through an area of bone spicule pigmentation. There is a line of high signal arising from the inner retina that corresponds to the known location of pigmented RPE cells in these lesions. There is also marked attenuation of signal from deeper tissues. (B) Macular island in RP. The OCT scan on the right corresponds to the location of the white line, (x y), on the photograph on the left. This spans an area of relatively normal retina on the left and a hypopigmented region to the right. The outer band thickness is much greater in the latter region. (C) Retinal laser lesion. The OCT scan on the right corresponds to the location of the white line,(x y), on the photograph on the left. This passes through a single laser lesion with a pale rim. The inner band has higher signal in the region of the center of the lesion, and the outer band is thicker at the pale rim. Vertical white bar marker, 250 m. therefore, the thinner it appeared. This may have been caused either by the loss of coherence with multiple scattering of light, 23 resulting in the detection of low interferometric signal, or by absorption and scattering of light so that less was returned to the OCT detector. Overall, therefore, the z-axis resolution of OCT within poorly scattering tissues may be limited. In the retina, change of refractive index may be responsible for the inner aspect of the inner band, and scattering could define its outer border. On entering the OCT detector system, light that was returned from the tissue being examined was allowed to interfere with that from the reference arm. The resultant intensity of this interference was recorded at 500 points along an individual z-axis scan and was represented by a logarithmic pseudocolor scale. The arbitrary units of the signal intensity itself had a range of 0 to 1600, whereas the pseudocolor scale consists of 16 colors. This degradation of signal resolution is in contrast with other imaging systems in which attempts at improving image resolution have involved extending gray scale. The result of the pseudocolor scale is likely to be a grouping of a wide range of higher signal intensities into single-color bands of red or white. Significant variations in signal intensity within the inner and outer bands of retinal OCT images thus may not be displayed, with a consequent loss of spatial resolution in the z-axis. When imaging retina, the effect may be to compress a 13-layered system 13 into an image with just three or four bands. Postprocessing of OCT images may also limit resolution. The z-plane is always displayed in the vertical axis, and each pixel has a depth measurement of 4 m. The spatial resolution of OCT images in the x y plane is dependent on both the optical limitations of the ocular media and the design constraints of the instrument. These limitations are apparent when attempting to image over a wide x y field. Each OCT image is constructed from 100 equally spaced individual z-axis scans, the x y dimensions of each being constant, irrespective of the area of retina scanned. Thus, the greater the x y field imaged, the greater the spacing between the individual z-axis scans, ranging from 10 to 110 m, and the lower the resolution in the x y plane. 12 Thus, for example, when the longest scan is used, the foveola may be missed, because it is only 150 m across. It is therefore evident that, in all three dimensions, the pixelation of images gives rise to both a mismatch between theoretical and practical resolution and indistinct borders on images of tissues. The appearance of a recorded image is also dependent on the preference of individual operators. Although the focusing

9 2340 Chauhan and Marshall IOVS, September 1999, Vol. 40, No. 10 of an image is readily and accurately repeatable and the power setting can be standardized, polarization is a variable that cannot be quantified on current scanners, and settings vary among observers. In this study, by examining raw images, we have shown polarization to have a significant effect on the measurement of some parameters, both among observers and observations by a single operator. The thickness of the inner band, for example, varied by up to 30% among observers and had a maximum of 18% for a single observer. The variances of inner and outer band thickness measurements were least for images that had undergone gaussian smoothing. For measurements of the total retinal thickness, however, there was no significant difference among raw, normalized, median-smoothed, and gaussian-smoothed images. Given the difficulties in resolving the various components of the OCT image, it is surprising that only one other study has been undertaken to correlate OCT images with histology. Toth et al. 14 assumed that the innermost aspect of the inner highsignal band corresponded to the inner limiting membrane and that the innermost aspect of the outer band corresponded to the apical surface of the RPE. They then compressed the histologic images, by between 4% and 12%, so that the distance between these two components matched those of their OCT scans. They reported that the outer aspect of the inner band was coincident with the outer aspect of the RNFL. Thus their images had four high-signal bands that corresponded to the RNFL, IPL, the residual element of the fiber layer of Henle, which they erroneously described as the OPL, and the RPE. In all the high-signal areas identified by Toth et al. 14 subcellular components are present, with dimensions between 1 and 3 m, which are potentially high scatterers of light. In the low-signal areas the nuclei have dimensions between 4 and 9 m. 14 Although their experimental findings seem to indicate a direct correlation between OCT strata and retinal cellular components, the match is only superficial. For the bands said to correspond to the RNFL and OPL, the ranges of thickness measurements were 3.3 and 7.9 times greater on OCT than light microscopy for the published images. Thus, if, for example, the furthest excursion of the inner band were taken as the true extent of this layer, it would include both the RNFL and 80% of the GCL. Although the previous study did not state that image postprocessing programs had been used, the images would suggest that they had. It is not clear which program was used, however, and we have used the manufacturer s gaussian smoothing program in our study, because this resulted in the lesser variance in measurements (coefficient of variance, 0.08; cf for median smoothing). Although Toth et al. 14 concluded that there was a close correlation between the OCT bands and specific retinal layers, such as the RNFL, our findings failed to support this completely. In particular, the persistence of an inner band after the deliberate destruction of inner retinal layers contradicts the notion of tissue-specific signal. Toth et al. 14 assumed that the inner border of the outer band corresponded to the apical RPE, and their histologic images were manipulated to generate a match in total retinal thickness. In our study, no image manipulation was required to generate matches in retinal thickness, because they could be directly correlated with ablated samples. Direct evidence was determined, locating the inner border of the outer band at the level of the RPE. Bone spicules are formed in RP by the migration and aggregation of pigmented RPE cells surrounding retinal capillaries. Although their precise location cannot be determined on biomicroscopy for direct comparison with OCT, their abnormal retinal distribution is well established. 17,24 The high signal corresponding to the location of these pigmented RPE cells was strongly suggestive of a role for this cell in OCT signal generation. In contrast to this situation of high melanin concentration, the RPE cells at the pale rim of a retinal laser lesion differ from RPE cells in normal nearby retina, in that they have little or no melanin within them. The corresponding loss of signal intensity in the outer band therefore demonstrates that it was not RPE cells that were responsible for the generation of signal but their subcellular components such as melanin and lipofuscin. It follows that the innermost aspect of the outer band is likely to correspond to the apical region of the RPE layer, because it is within this 3- m layer that melanin granules are concentrated. This was confirmed in our in vitro study by the observation that the location of the inner border of the outer band remained constant and close to the RPE, even with progressive ablation. Another significant optical effect of melanin seen in these studies was an attenuation of signal from deeper tissues. The most notable examples were seen with bone spicules when the signal from deeper tissues was almost absent. Further, the apparent unveiling of signal from tissues deep to focal regions of RPE hypopigmentation is also consistent with a downstream signal-attenuating role for melanin. The candidate optical phenomena responsible for the duality of high signal generation and downstream signal attenuation are absorption, reflection, and scatter. Absorption is unlikely to account for the downstream signal attenuation, because only 3% of incident light at a wavelength of 830 nm is absorbed by RPE melanin. 25 Similarly, the degree of attenuation seen deep to melanin cannot be caused by reflection of light by melanin, because a maximum of only 4% of light is reflected by polished interfaces in optical instruments. 26 Scatter is the only optical phenomenon that can be responsible for both observations, and melanin is already known to be a strong scatterer of light. 27 The role of scatter in the attenuation of downstream signal is likely to have two components, each resulting in a significant reduction in the light returned to the detector. The first is the deviation of light away from deeper tissues so that less is incident on them. Second, any light that reaches deeper tissues and is returned toward the detector may be further scattered by the melanin and therefore deviated away from the detector. The clinical significance of this phenomenon is that diseases affecting tissues deep to the RPE may not be directly demonstrable with OCT. Changes relating to choroidal neovascularization have been described elsewhere and are limited to a description of the thickness and irregularity of the outer band. 28 Changes of this nature were observed in our study of laser lesions and, although the effect of RPE pigmentation was noted in the previous study, the possible influence of this on their findings was not discussed. 29 The influence of tissues on the signal from deeper tissues is not exclusive to melanin and was seen with the progressive ablation of tapetal retina. In this case, the thickness of the outer band increased linearly with progressive ablation of overlying retinal tissue, presumably because of a loss of attenuation from those tissues. That this was not significant with nontapetal or human retina until after ablation of the RPE may be accounted