Carlos Alexandre de Amorim Garcia Filho, Zohar Yehoshua, Giovanni Gregori, Carmen A. Puliafito and Philip J. Rosenfeld

Physical principles of optical coherence tomography

During the past two decades, optical coherence tomography (OCT) has become an essential tool in ophthalmology. Its ability to image detailed ocular structures noninvasively in vivo with high resolution has revolutionized patient care.1,2 OCT technology is based on the principle of low-coherence interferometry, where a low-coherence (high-bandwidth) light beam is directed on to the target tissue and the scattered back-reflected light is combined with a second beam (reference beam), which was split off from the original light beam. The resulting interference patterns are used to reconstruct an axial A-scan, which represents the scattering properties of the tissue along the beam path. Moving the beam of light along the tissue in a line results in a compilation of A-scans with each A-scan having a different incidence point. From all these A-scans, a two-dimensional cross-sectional image of the target tissue can be reconstructed and this is known as a B-scan.

Typically OCT instruments use an infrared light source centered at a wavelength of about 840 nm. For a given wavelength, the axial resolution is dictated by the bandwidth of the light source. The latest commercial instruments typically have an axial resolution of approximately 5 µm, while research instruments have been built with a resolution as high as approximately 2 µm.¹ The lateral resolution is limited by the diffraction caused by the pupil and it is normally about 20 µm. For clinical purposes, the image acquisition time is limited by the patient’s ability to avoid eye movements, i.e., less than 2 seconds in the typical patient. The instrument’s scanning speed (number of A-scans acquired per second) is then the crucial parameter determining the amount of data available for a single OCT dataset.

The early OCT instruments, known as time domain OCT (TD-OCT), used a single photo detector, and an A-scan was created by moving a mirror to change the optical path of the reference beam in order to match different axial depths in the target tissue. This setup limited the scanning speed to a few thousand A-scans per second. A newer technique, known as spectral domain OCT (SD-OCT), Fourier domain OCT (FD-OCT), or high-definition OCT (HD-OCT), is able to acquire an entire A-scan by using an array of detectors instead of using multiple reference beams from a moving mirror. Scanning speeds with SD-OCT instruments can exceed 100 000 A-scans per second, about 200 times faster than TD-OCT. Currently available SD-OCT commercial systems operate at a scanning rate of approximately 27 000 A-scans per second.¹

The scanning pattern with the commercial TD-OCT instrument (Stratus OCT, Carl Zeiss Meditec, Dublin, CA) incorporated six radial, concentric, 6-mm-long B-scans centered on the fovea. With the recent development of high-speed SD-OCT systems, several novel and important imaging strategies have been introduced based on acquiring three-dimensional datasets and B-scan averaging (Fig. 3.1).

Three-dimensional datasets are obtained using a dense two-dimensional raster array over a relatively large retinal region. The resulting datasets can be rendered as a volume image in three dimensions and can be analyzed by showing two-dimensional slices (i.e., sequences of parallel B-scans). Three-dimensional datasets give detailed information about the retinal structure over large areas. In addition, it is possible to generate en face fundus-like images directly from the OCT datasets. These OCT fundus images (OFIs) provide an accurate spatial colocalization of retinal features observed on the en face and cross-sectional images. Therefore, exact correlations can be achieved between the retinal cross-sectional geometry seen on the OCT B-scans and the retinal landmarks seen on en face images, known as the OFI. The potential exists for registration between several SD-OCT datasets of the same eye and images obtained using other imaging modalities, such as color fundus photography, fluorescein angiography, and fundus autofluorescence imaging. This holds the promise for an unprecedented ability to describe and monitor changes in the local geometry of the retina.³ In addition to the OFI generated by a full OCT dataset, partial OFIs (or slabs) can be generated to produce en face renderings that correspond to particular retinal layers or features.⁴ These slabs can be very useful to visualize and quantify specific pathologies (Fig. 3.2).

The scanning speed of SD-OCT can also be used to produce very high-quality individual B-scan images through a combination of high sampling density and image averaging. One of the main factors affecting the perceived quality of OCT images is noise, in particular the speckle noise which is responsible for the characteristic “granular” appearance of OCT. Noise can be reduced through the acquisition, registration, and averaging of a number of B-scans at approximately the same retinal position (Fig. 3.3).

Although en face registration and B-scan averaging strategies can be implemented in many ways, a particularly powerful and flexible solution is the use of a built-in laser eye-tracking system. The main limitation of this approach is that the necessary acquisition times can become very long and sometimes unmanageable, particularly for large raster scans and for subjects with poor fixation.

Recently, different companies have invested in research in the field of retinal imaging, especially in the development and improvement of SD-OCT. It is not an objective of this section to discuss the differences between each of the currently available instruments since these instruments are continuously evolving. Table 3.1 lists currently available instruments.

Quantitative analysis of OCT datasets

A crucial step towards a clinically useful, quantitative understanding of the retinal anatomy is the development of accurate, robust, reproducible segmentation algorithms that can automatically identify the boundaries between specific retinal layers and/or other retinal features. The currently available SD-OCT instruments have several advantages over the previous generation of TD-OCT instruments. SD-OCT instruments generally have a higher axial resolution and can produce B-scans with better image quality by increasing the A-scan density and through averaging techniques. Much more importantly, the higher scanning speed made possible by the SD-OCT technology reduces the effect of artifacts associated with eye motion and produces images that provide a true picture of the retinal geometry. The large, dense raster scans make it possible to obtain detailed surfaces of individual retina layers over large areas, resulting in segmentation maps. These maps allow for an unprecedented visualization and quantitative evaluation of the corresponding retinal structures.

Several commercially available SD-OCT instruments offer some level of quantitative analysis using different, proprietary segmentation algorithms. The various segmentation algorithms make different design choices and have been shown to have very different performance profiles in terms of accuracy, reproducibility, and robustness.5–8 Care should be exercised when comparing measurements obtained from different OCT instruments.

The most commonly used quantitative parameter derived from OCT datasets is retinal thickness, obtained by segmenting the internal limiting membrane (ILM) and a boundary representing the retinal pigment epithelium (RPE). This information can be used to generate surface maps of the ILM and the RPE as well as two-dimensional and three-dimensional retinal thickness maps. These maps can be very useful in identifying and describing deviations from the normal anatomy and changes over time. Registering OCT datasets acquired over time can give very precise information about the dynamics of disease progression and response to treatment based on changes in retinal anatomy (Fig. 3.4).

It is important to keep in mind that there is some confusion in the definition of the outer retinal boundary. In a normal eye, the bright reflective band at the external aspect of the retina, often referred to as the RPE complex, can be resolved in ultrahigh-resolution images, and occasionally in images acquired with a commercially available SD-OCT instrument, consisting of three individual layers.⁹ Different segmentation algorithms from different instruments tend to follow different edges and therefore result in different measurements. For example, the Spectralis SD-OCT instrument typically follows the posterior edge of the RPE complex, the Stratus TD-OCT instrument typically follows the inner segment–outer segment (IS/OS) junction, which is anterior to the RPE complex, and the Cirrus SD-OCT instrument typically follows the anterior edge of the RPE layer (Fig. 3.5). This situation becomes even more complicated and sometimes inconsistent when the normal retinal structure is deformed by the presence of pathology.¹⁰

In addition to total retinal thickness, a number of other quantitative parameters have been proposed. For example, it is possible to obtain measurements of particular retinal layers, such as the thickness of the ganglion cell layer or the thickness of the photoreceptors’ outer segments, as well as measurements of retinal lesions, like the area of geographic atrophy (GA).9,11,12

An area of particular promise is the measurement of RPE deformations associated with drusen.13,14 These measurements are obtained by comparing the actual RPE geometry with the geometry of a virtual RPE free of deformations. Parameters like drusen area and volume can be generated in a fully automated manner and have been shown to be quite robust and reproducible (Fig. 3.6).

The amount of information provided by each dataset, together with the possibility for image registration and longitudinal studies, makes SD-OCT a very promising new tool for the quantitative study of retinal pathologies. These capabilities solve the biggest problems associated with the Stratus TD-OCT retinal thickness maps: the lack of precise correspondence between the B-scans and the retinal topography, the difficulties associated with eye movements, and the need for significant interpolation of the data due to undersampling of the retina. Despite the advantages of SD-OCT, segmentation algorithms can produce artifacts, particularly in the presence of macular disorders with complex morphology like neovascular age-related macular degeneration (AMD).15–19 Therefore, it is important to be vigilant and monitor the quality of the segmentation in order to eliminate artifacts arising from flawed segmentation and associated measurements.

Normal macular anatomy

The OCT image closely approximates the histological appearance of the macula and, for this reason, it has been referred to as an in vivo optical biopsy. With the increase in the axial resolution of the new SD-OCT instruments (5–8 µm) and the ultrahigh-resolution OCT (2 µm), it has become possible to correlate OCT images accurately with histological features of the retina.²⁰ However, care must be taken when making assumptions about these correlations because histological sections require fixation and exogenous staining to produce contrast within tissue, and this can introduce artifacts, while OCT relies on intrinsic differences in tissue optical properties to produce image contrast.²¹ When light travels through the retinal tissue it can be reflected, scattered, or absorbed, and this creates the multilayered pattern of the retina. The angle of incidence of the light, motion artifacts, speckled noise, and image contrast can affect the axial resolution of retinal imaging. Therefore, one-to-one correspondence of histology with OCT images cannot be expected.^2,21

Although the interpretation of features of the inner retina, which can be defined for our purpose to span from the ILM to the junction of the inner and outer segments of the photoreceptors, appears to correlate well with histology, the OCT features of the outer retina are less well understood and remain a topic of discussion (Fig. 3.7).^9,22,23

The first detected layer in most OCT scans is the ILM that appears as a hyperreflective layer at the vitreoretinal interface. In some patients, the posterior hyaloid can be seen above the ILM as a hyperreflective layer. Within the retina, the retinal nerve fiber layer and the plexiform layers (both inner and outer) are seen as hyperreflective while the ganglion cell layer and the nuclear layers (both inner and outer) are hyporeflective. A recent study demonstrated that the incidence of the light beam could affect the appearance of Henle’s fiber layer by OCT, resulting in a thin hyperreflective layer corresponding to the photoreceptor synapses or a thicker hyperreflective layer corresponding to photoreceptor axonal extensions enveloped by the outer cytoplasm of Müller cells (Fig. 3.8).²⁴ The retinal vessels may sometimes be seen on OCT images as circular hyperreflective structures located in the inner retina, with a vertical shadow or reduced reflectivity extending into deeper layers.

Outside the central fovea, commercially available SD-OCT instruments typically resolve four bands in the outer retina. There is discordance between different authors regarding which anatomical structure correlates with each band.20,21,25 The innermost band has been attributed to the external limiting membrane (ELM). This band is typically thinner and fainter than the others. The nomenclature for the middle two bands has much less supportive evidence. The second of the four bands has been commonly ascribed to the boundary between the IS/OS of the photoreceptors and the third band is referred to as either the OS tips or as Verhoeff’s membrane.^9,26 A recent study suggested that the second band was the ellipsoid section of the photoreceptors (inner segment) instead of the IS/OS junction and that the third band appears to correspond to the contact cylinder between the RPE apical process and the external portion of the cone outer segment. This band typically merges with the fourth band in the central fovea and this is explained by a greater height of the contact cylinder of the cones and RPE outside the fovea.²⁵ The fourth hyperreflective outer retinal band is attributed to the RPE, with potential contribution from Bruch’s membrane and choriocapillaris, with abundant experimental and clinical evidence supporting this designation.^9,22,27

Although the current SD-OCT uses a short wavelength of approximately 840 nm, which results in light scattering at the level of the RPE and a lower signal from the deep choroidal tissue, it is also possible to image the choroid and extract quantitative information (Fig. 3.9).^28–31 Choroidal thickness may be influenced by age, axial length, and perhaps refractive abnormalities. It also varies in different retinal regions within the same normal subject, being thickest beneath the fovea,³¹ or in the superior outer macula (Early Treatment Diabetic Retinopathy Study (ETDRS) subfield), with the thinnest choroid being located in the nasal outer ETDRS subfield.³² When centering the optic nerve head as a reference point, the choroid appears thin in the peripapillary region and increases in thickness with eccentricity in all directions, up to a certain point, except inferiorly.³² This is the embryonic location of the optic fissure closure and thus may be responsible for the localized thinning.^32,33 High-penetration OCT uses a light source with a wavelength around 1050 nm that allows a better visualization of the posterior choroid and sclera than currently available SD-OCT instruments.^30,33,34

The high axial resolution and the different scan patterns offered by SD-OCT provide comprehensive structural information that can be used to map retinal layer thicknesses and perform volumetric analyses. Using different SD-OCT instruments, several authors have reported an approximate central retinal thickness of 265 µm in normal subjects.³⁵ However, caution is required, as errors in automated measurements may occur and are more often found in macular disorders with complex morphology like neovascular AMD, which alters the ability of segmentation algorithms to detect normal boundaries.^15–19 Therefore, care must be taken that high-quality and artifact-free scans are obtained before running the retinal thickness algorithm.

It is essential to consider the following points when analyzing OCT images: location, shape, and reflectivity of the structure, along with its histological correlation. It is also important to remember that the alignment of the instrument with the pupil can generate signals that may lead to a misinterpretation of the exam.

Characteristic OCT findings in several common retinal disorders which are frequently studied using OCT are discussed below. For OCT findings, in other disorders, such as retinal degenerations, the reader is directed to the specific chapters describing these diseases.

SD-OCT in retinal disorders