1. Introduction

Optical coherence tomography (OCT) has revolutionized our ability to image the retina and choroid, but the development of accurate and automated algorithms to measure the retinal and choroidal layers, as well as other features, have lagged behind the technical advancements that have led to improvements in image quality [1–5]. With increased imaging speed, scanning density, averaged B-scans, and longer wavelengths, OCT imaging now allows for wider and deeper scans with improved image quality of the retina and choroid. However, accurate algorithms that segment anatomic layers beyond those layers that display a distinct relative signal intensity, such as the internal limiting membrane and retinal pigment epithelium (RPE), remain an ongoing challenge for OCT developers and manufacturers. Deep learning algorithms have been a useful alternative strategy for improving the segmentation of retinal and choroidal layers, but these strategies are unique for specific instruments and scan patterns and cannot be easily used across platforms without additional intensive supervised retraining [6–13]. However, even these strategies have their own limitations in identifying subtle structures, particularly those associated with decreased signal intensity at greater depths within the choroid.

The need for improved OCT segmentation strategies is particularly important for the study of age-related macular degeneration (AMD). Our understanding of AMD, the leading cause of irreversible vision loss and blindness among the elderly worldwide [14], has relied traditionally on the use of color fundus imaging (CFI), fundus autofluorescence imaging (FAFI), and intravenous dye-based angiographic imaging using fluorescein and indocyanine green; however, more recently, OCT and OCT angiography (OCTA) imaging have become the gold-standards for diagnosing and following disease progression in AMD, both in the clinics and in clinical trials [15,16]. Since AMD causes both anatomic and angiographic changes in multiple layers throughout the macula, OCT is ideally suited to detect and monitor these changes in the retina, the retinal pigment epithelium (RPE), Bruch’s membrane (BM), choriocapillaris (CC), and choroid [17]. Based on fundus exam and color fundus imaging, AMD has been classified into three progressive stages that can be easily translated into OCT-defined stages [18] for which the ability of OCT to image choroid is required.

OCT imaging also allows for an exploration of the choroid, and the choroid presents some unique challenges given the progressive loss of signal strength associated with tissue depth, partly due to 1) technical limitations in the data sampling of the OCT signal leading to signal sensitivity falling off along the depth and 2) inherent optical property of biological tissue leading to progressive signal attenuation along the depth [19,20]. One strategy to mitigate the data sampling issue was to employ enhanced depth imaging (EDI) developed for use with the Heidelberg Spectralis spectral-domain OCT (SD-OCT) instrument that operates at a wavelength of 870 nm [21,22]. The method of EDI involves moving the OCT instrument closer to the eye, leading to the zero-delay line being placed deeper into the tissue and closer to the choroid, which increases the signal fidelity sampled at deeper structures, therefore enhancing the sensitivity of choroidal imaging. However, the EDI strategy does not address the fact that, at the wavelength of 870 nm, the light penetration into the choroid is limited by the relatively high optical scattering at and below the RPE complex. Because optical scattering strength in this region decreases with the increase of wavelength in the near infrared region [23], the development of longer wavelength SD-OCT and swept-source OCT (SS-OCT) [24–26] are the preferred technical alternatives for delivering improved choroidal imaging. With the OCT instruments operating at a wavelength of about 1050 nm, the EDI strategy was no longer needed [27–29]. However, even with longer wavelengths, the optical scattering property of the RPE and choroidal tissues still progressively attenuates the OCT signal strength below the retina, a serious problem for the development of intensity-based segmentation algorithms of the choroidal-scleral boundary [30,31].

In this perspective, we will first introduce the staging of AMD using OCT imaging and then show how the combination of optical attenuation coefficients (OACs) with OCT imaging enhances the detection of retinal and choroidal anatomy. We will describe how OAC improves the detection of the choroidal-scleral boundary and the choroidal vasculature, which greatly facilitated measurements of choroidal thickness and the choroidal vascularity index (CVI). In addition, we will describe how the use of depth-resolved OAC imaging in conjunction with conventional multi-layer segmentation strategies improves the detection of anatomic layers, helping with the clinical staging of AMD and the ability to predict disease progression.

2. Stages of AMD and the role of OCT imaging

The early and intermediate stages of AMD are characterized by the presence of drusen, which appear as yellow focal deposits in the macula that represent the accumulation of lipoproteins between the RPE and BM and correspond to focal RPE detachments [18]. Early AMD is defined by the size of these drusen. When the en face greatest linear dimension (GLD) of drusen on CFI measures ≤ 125 µm in the absence of hyperpigmentation or when the GLD is < 62 µm in the presence of hyperpigmentation, these eyes are classified as the early stage of AMD. The intermediate stage AMD is defined by the presence of drusen with GLDs ≥ 62 µm in the presence of hyperpigmentation or drusen with a GLD ≥ 125 µm. OCT imaging is ideally suited to both replace and refine the staging system based on CFI since OCT imaging enables the three-dimensional (3D) imaging of these drusen [32–34]. Algorithms capable of segmenting the RPE and BM can provide accurate and reproducible volume and area measurements of these drusen. Moreover, the presence of hyperpigmentation within the retina and along the RPE can be imaged as focal areas of hyperreflectivity using OCT [35–37]. However, up until recently, the quantitation of hyperpigmentation has been a challenge, which has been overcome by the use of OCT algorithms that incorporate neural networks [38,39] or OACs [40].

The late stages of AMD are characterized by the formation of geographic atrophy (GA) or the formation of macular neovascularization (MNV) [18]. GA, a form of macular atrophy identified in AMD, is characterized by the loss of the outer retina, RPE, and choriocapillaris. Historically, the formation of atrophy was diagnosed by the appearance of whitish lesions on fundus exam or CFI that correspond to absence of the RPE and visualization of the underlying sclera. Such lesions appear dark on FAFI due to the absence of the RPE. OCT is now the preferred imaging strategy for the diagnosis of macular atrophy since the absence of the affected anatomic layers can be directly visualized [41,42]. Macula atrophy can also be inferred using OCT imaging by the appearance of choroidal hypertransmission defects (hyperTDs). In areas where the RPE is attenuated or absent, light penetration into the choroid and sclera increases, and so does the corresponding back-scattering, resulting in increased brightness from these layers on B-scans and en face OCT images [41,42]. Different OCT B-scan based definitions have arisen to characterize these areas of impending or total atrophy, including nascent GA, incomplete outer retina and RPE atrophy (iRORA), and complete RORA (cRORA) [42]. Another strategy to identify these atrophic lesions relies on a combination of both en face and B-scan images from dense OCT volumetric scans to identify regions known as choroidal hyperTDs [43–48]. OCT algorithms have been developed to diagnose these regions of atrophy, and these algorithms have been improved by incorporating OACs into the OCT images [10].

MNV, another late stage of AMD, is characterized by the ingrowth of neovascularization from the CC through BM into the subRPE space (Type 1 MNV) and from the CC through BM and through the RPE into the sub-retinal space (Type 2 MNV) [49]. In addition, MNV can arise from the deep capillary plexus of the retina and extend into the outer retina with further extension through the RPE to BM (Type 3 MNV) [49]. Historically, these neovascular lesions were diagnosed using dye-based angiography (FA and ICGA), with FA particularly sensitive to the presence of exudation. However, the presence of MNV is now routinely diagnosed using both structural OCT and OCT angiography (OCTA) [49,50]. Structural OCT back-scattering identifies back-scattering exudation by the presence of intraretinal and subretinal fluid and can identify the nonexudative type 1 MNV by the presence of a low-lying double layer sign, a shallow irregular RPE detachment characteristic of type 1 MNV [51,52]. OCTA is particularly useful for identifying and confirming the presence of these non-exudative forms of MNV, which arise prior to the onset of exudation, and these OCTA non-exudative neovascular lesions identify those eyes at risk for developing exudation. These non-exudative lesions also result following treatment with vascular endothelial growth factor (VEGF) inhibitors [53–56]. OCT imaging has proven to be an invaluable imaging method for validating and improving the current staging system for AMD by combining both anatomic and angiographic features of AMD that had previously required the use of multimodal imaging platforms. In addition, OCT imaging has allowed for the exploration of the choroid, a layer vital to the health of the macula and never characterized in AMD using the other imaging modalities.

3. Choroidal imaging with optical attenuation correction OCT imaging

Zhou et al. [30,31] were the first to use attenuation correction to visualize the choroid and reliably segment the scleral-choroidal boundary from SS-OCT scans. Optical attenuation correction is ideally suited for visualizing deeper choroidal layers in which the OCT signal intensity is significantly attenuated. This approach has advantages compared with algorithms that relied on conventional signal-intensity SS-OCT images. This allowed for the creation of reliable choroidal thickness maps (Fig. 1) that are clinical useful in the investigations of choroidal involvement in ocular diseases. Attenuation correction relies on the use of OACs, which are an inherent property of the local tissue components at an operating OCT wavelength [57]. These OACs are not dependent on the depth information, thus do not vary between and within anatomic layers at different depths within OCT images. Therefore, if leveraged in the development of algorithms, these OACs would help improve deeper layer detection [57,58]. As a consequence, using attenuation correction enhances the signal strength throughout a given tissue, such as the choroid, so that the contrast in the deep choroidal layers is increased and structures within the choroid and at the choroidal-scleral boundary can be more easily segmented [30]. Another strategy is to replace OCT signal intensities with depth-resolved OAC values since this can accentuate the differences between retinal layers. Again, the result can be a better visualization of the layers and therefore better segmentation [57].

 

Fig. 1. Improved segmentation of the choroidal-scleral boundary and associated choroidal thickness maps after attenuation correction. (A,D) Manually segmented choroidal thickness maps and B-scans from attenuation corrected OCT scans. (B,E) Automatically segmented choroidal thickness maps and B-scans from regular OCT scans. (C,F) Automatically segmented choroidal thickness maps and B-scans from attenuation corrected OCT scans. (Figure reprinted with permission from Zhou et al. [30] in Biomedical Optics Express). White scale bar = 500µm

Download Full Size | PDF

Attenuation correction not only facilitated reliable segmentation of the choroidal-scleral boundary, but the choroidal vessels as well, which resulted in visualization of the major choroidal vessels and measurement of the choroidal vascularity index (CVI) within the choroid (Fig. 2). Using volume measurements of both the choroid and the choroidal vessels from dense 12X12mm OCT raster scans, we calculated the CVI, which was originally defined as the ratio of the vascular lumen area in the choroid to the total area of the choroid on OCT B-scans, but now re-defined as the ratio of the choroidal vessel volume measurements to the total volume of the choroid.

 

Fig. 2. Improvement of choroidal vessel visualization using attenuation correction. (A-B) Representative B-scan before and after attenuation correction. Shadows from retinal vessels shown with yellow arrows were markedly reduced after attenuation correction. (C) Lateral pixel intensity profiles along the shadows (indicated with red and blue arrows) showed the percentage difference from the mean intensities before and after attenuation correction. After attenuation correction, the indicated shadows were successfully eliminated. (D-E) Minimum projection of choroidal vessels of a normal eye without (D) and with (E) attenuation correction. (F) Magnified regions (red and blue squares) of the vasculature showed the elimination of artifacts from the retina after attenuation correction. (Figure reprinted with permission from Zhou et al. [30] in Biomedical Optics Express). White scale bar = 1 mm.

Download Full Size | PDF

Using 12X12mm volumetric scans from a normative database of subjects ranging in ages from their 20s to their 80s, Zhou et al. [31] showed that choroidal thickness was greatest in the macula and thinnest inferior to the optic nerve, and this thickness progressively decreased in all regions with increasing age. They were able to establish 95% normal limits for choroidal thickness in each decade of life. While the choroidal thickness decreased with age, the CVI was found to be stable with age. Using the same choroidal attenuation correction strategy, Shi et al. [59] showed that in eyes with AMD and geographic atrophy (GA), decreased baseline CVI measurements within the area of atrophy correlated with the annual growth rate of GA [59]. They also showed that the overall CVI measurements were unusually low for any given choroidal thickness measurement in these eyes with GA. In addition, Shen et al. [60] showed that in eyes with polypoidal choroidal vasculopathy (PCV) treated with intravitreal vascular endothelial growth factor (VEGF) inhibitors, the anti-VEGF treatment caused a decrease in choroidal thickness, but an increase in CVI, which suggested that the PCV lesions served as a high-flow vascular shunt between the choroidal arterial and venous systems prior to anti-VEGF therapy and the resorption of excess choroidal transudation after the treatment. The same algorithm was also used to study choroidal changes in eyes with nonexudative AMD prior to the onset of exudation to determine if choroidal changes predicted the onset of exudation, but no predictive changes were identified [61]. In eyes with proliferative diabetic retinopathy, choroidal thickness was found to decrease after treatment with panretinal photocoagulation in the regions where the laser was applied, but the CVI was unchanged [62]. Additional imaging studies investigating the impact of anti-VEGF therapy on choroidal anatomy and blood flow are ongoing.

4. Identification and quantitation of choroidal hypertransmission defects

Vermeer et al. [57] first proposed a depth-resolved single scattering model for assigning pixel dependent OAC values at different OCT A-scan depths, and this method of calculating these OAC values at each pixel at different depths using SS-OCT scans was adopted by Zhou et al. [30,31] and Chu et al. [10] to develop novel segmentation algorithms. This strategy of replacing conventional signal intensities with OAC values further accentuates highly reflective layers and allows for greater discrimination between adjacent layers, which we have used to develop automated segmentation strategies for use with SS-OCTA scans. This strategy was shown to be particularly useful for imaging the RPE and identifying where the RPE is attenuated or lost for the diagnosis of choroidal hypertransmission defects (hyperTDs) [47,48].

The presence of hyperTDs is commonly associated with areas of geographic atrophy (GA) in AMD, and these areas correspond to the loss of the RPE and the photoreceptor layer, which is characteristic of late stage nonexudative AMD. Historically, GA was diagnosed based on fundus biomicroscopy, CFI, or AFI, but the Classification of Atrophy Meeting (CAM) group recently recommended that OCT be used as the preferred imaging modality for GA and further developed new terminology to define regions of atrophy [41,42]. The CAM group defined complete RPE and outer retinal atrophy (cRORA) as the preferred OCT imaging term to replace GA. cRORA is defined based on OCT B-scans and requires a choroidal hyperTD measuring at least 250 µm in the B-scan plane as well as outer retinal and RPE attenuation. When the hyperTD in the B-scan plane measures less than 250 µm we have incomplete RORA (iRORA).

Another strategy for identifying regions with hyperTDs is to identify lesions on en face OCT slabs positioned 64-400 µm below BM derived from volumetric OCT scans comprised of closely spaced B-scans, ideally so that the en face image has nearly uniform isotropic spacing between the A-scans. This slab is commonly known as the subRPE slab. Areas of hyperTDs appear as bright spots on the subRPE en face image. In a natural history study, it was shown that once the greatest linear dimension (GLD) of a bright spot reaches at least 250 µm, these bright spots persist over time, so they were designated as persistent choroidal hyperTDs. Persistent hyperTDs are easily identified and measured using the en face slab [48]. Chu et al. [10] developed an automated algorithm for the identification and measurement of hyperTDs, as well as the measurement of drusen areas and volumes. This algorithm uses the information from the subRPE intensity slab, the OAC values map along the RPE layer, and the OAC elevation map measuring the distance between the RPE and BM, to segment the hyperTDs. This algorithm is ideal for following disease progression in AMD and can be used in clinical trials to detect progression from intermediate AMD to nonexudative late AMD with macular atrophy [63]. (Fig. 3)

 

Fig. 3. Geographic atrophy (GA) visualized using the optical attenuation coefficient (OAC) estimated from SS-OCT images. Representative images are shown from a normal eye (A – H, 34 years old male) and an eye with GA eye (I – P, 76 years old female), respectively. A, I: OCT subRPE images. B, J: ground truth of GA generated by graders. C, K: composite OAC false color images. D, L: OAC elevation maps. E, M: OAC max images. F, N: OAC sum images. G, O: OCT B-scans with its location represented by dashed lines in panels A and I. H, P: corresponding OAC B-scans with the same location of G and O. All images are from 6 × 6 mm SS-OCT scans. All B-scans were flattened based on BM segmentation. (Figure reprinted with permission from Chu et al. [10] in Biomedical Optics Express)

Download Full Size | PDF

5. Imaging of drusen and basal laminar deposits with OAC-defined OCT elevation maps

Another example of an improved OAC depth-resolved segmentation algorithm is one capable of identifying the subtle separation between the RPE and BM created by basal laminar deposits (BLDs) [64]. Intensity-based OCT algorithms to measure drusen area and volume have been available and these algorithms identify the elevations of the RPE layer above a baseline analogous to BM [32,33,65,66]. More recently, a reliable segmentation algorithm was developed to identify the BM with SS-OCT images [34]. While this RPE to BM segmentation algorithm identified most drusen, small drusen causing a separation between the RPE and BM of 20 µm or less would be missed. This type of low lying RPE elevation is often associated with BLDs and nonexudative type 1 MNV. By replacing the traditional segmentation of the RPE with the depth-resolved OAC values along the A-scan, Chu et al. [64] showed that the RPE layer could be better visualized and the separation between the RPE and BM caused by BLDs could be more easily measured. In addition, this algorithm that used the OAC-defined RPE and the conventionally defined BM was able to identify any RPE elevation such as a conventional druse, and these corresponding maps were referred to as OAC elevation maps (Figs. 3,4). Using this combined OAC/intensity-based algorithm, Chu et al. [64] mapped the BLDs around GA and showed that the baseline extent of the BLDs correlated with the growth rate of GA. These results were consistent with previous conventional OCT studies that associated the presence of these BLDs with the fastest growing GA lesions. However, Chu et al. were able to map these BLDs in a wide range of GA lesions and establish an association between BLDs and GA growth rates for all GA lesions. This algorithm identified a baseline risk factor that now helps clinicians identify those AMD eyes at greatest risk of progression.

 

Fig. 4. Images obtained using the optical attenuation coefficient (OAC) algorithms and the traditional swept-source OCT (SS-OCT) images with choroidal hypertransmission defect (hyperTDs) consistent with geographic atrophy. A: SS-OCT OAC B-scan with red dashed line depicting Bruch’s membrane (BM) B: OAC maximum projection en face image of the first slab showing the area of GA. C: OAC sum projection en face image of the first slab showing the area of GA. D: OAC B-scan with green dashed lines indicating the OAC identified retinal pigment epithelium (RPE) and the manually segmented BM. E: OAC elevation map showing the distance between the segmentation lines shown in panel D, with the associated color bar with a dynamic range of 0 to 100 µm. F: OAC false color composite image of panels B (red channel), C (green channel), and E (blue channel). G: The same SS-OCT B-scan image as in panel A, with yellow dashed lines depicting the subRPE slab from 64 µm below BM to 400 µm below BM. H: OCT sum en face projection of the subRPE slab depicting the area with choroidal hyperTDs. I: Manually labeled area of the geographic atrophy shown in panel H. (Figure reprinted with permission from Chu et al. [64] in American Journal of Ophthalmology)

Download Full Size | PDF

6. Identification and quantitation of hyperpigmentation in the retina and along the RPE

Hyperpigmentation in the macula has long been associated with both intermediate and late-stage AMD and has also been identified as a risk factor for disease progression [37,67]. Historically identified by fundus biomicrosopy and color fundus imaging, these foci of hyperpigmentation represent clumped RPE cells along the monolayer, with some escaping into the retina, and can be accompanied by non-RPE cells, likely inflammatory cells [36]. When these foci of hyperpigmentation occur in the retina, they are referred to as hyperreflective foci. However, these areas of hyperpigmentation can also occur along the RPE, often resulting from the proliferation and reduplication of the RPE. When the RPE cells escape from its monolayer and migrate into the retina, they give rise to hyperreflective foci, but investigators often neglect the hyperpigmentation in continuity with the RPE when quantifying the burden of pigment in the macula [35,38,67–69]. While OCT intensity-based algorithms have attempted to identify and quantitate the hyper-reflective foci in the retina, there had been no previous effort to identify the foci of hyperpigmentation along the RPE even though the foci in the retina and along the RPE both represent a continuum of RPE cells migrating from the monolayer into the retina [35,36,38,67–69].

Recently, Zhou et al. [40] developed a strategy to visualize and manually quantitate the presence of these foci of hyperpigmentation by using the same en face subRPE slab that is used to identify choroidal hyperTDs. Rather than identifying bright areas associated with choroidal hyperTDs, they identified dark areas known as choroidal hypotransmission defects (hypoTDs) that result when foci of hyperpigmentation scatter the laser light resulting in the absence of any signal from within the choroid [37]. The hyperreflectivity of these areas of hyperpigmentation not only cause the hypoTDs, but also provide a detection method using the depth-resolved OAC strategy. By developing an algorithm that combines the detection of a high OAC signal along the A-scan depth with the subRPE slab that detects the foci of hypoTDs under BM, Zhou et al. [40] were able to assess the total pigment burden in the macula and distinguish between intraretinal foci of hyperpigmentation and foci of increased pigmentation along the RPE (Fig. 5). This algorithm is currently being used to study eyes with AMD to assess whether the total pigment burden, or just the pigment burden in the retina, can serve as predictors of disease progression.

 

Fig. 5. Visualization of hyperpigmentation in the retina and along the RPE and their relationship with surrounding drusen. Side views (A-C), top view (D-F), and selected OCT B-scan views (G-I). The locations of the B-scans are indicated by the yellow dashed lines in (D-F) with hyperpigmentation in the retina highlighted in blue and hyperpigmentation along the RPE highlighted in red. (A, D, G) An eye without drusen but with both intraretinal hyperpigmentation and hyperpigmentation along the RPE. (B, E, H) An eye with drusen and hyperpigmentation along the RPE (highlighted in red) that appear as foci of hyperreflective material on the B-scan (H). (C, F, I) An eye with a central area of hyperreflective material along the RPE (highlighted in red). (Figure reprinted with permission from Zhou et al. [40] in Biomedical Optics Express). White scale bar = 1 mm.

Download Full Size | PDF

7. Measurement of outer retinal thickness

OCT algorithms have been developed for the measurement of photoreceptors (PR) and the inner segment/outer segment (IS/OS) band (also known as the ellipsoid zone [EZ]) on B-scans, but the reliability of these algorithms have not been validated in all stages of AMD [8,9,70–72]. The accurate segmentation of the EZ and PR layers need to be validated in eyes with features of AMD that disrupt the integrity of the IS/OS band, such as in the presence of reticular pseudodrusen, or features that alter the waveguiding properties of photoreceptors, as would be expected in the presence of RPE elevations, such as soft drusen [73]. Another published strategy attempted to segment the outer nuclear layer, specifically between the outer boundary of the outer nuclear layer and the outer plexiform layer, but this strategy is complicated by the changes in the reflectivity of Henle’s layer within the outer nuclear layer caused by the unique optical property of the elongated inner processes of the foveal cones and rods when the incident light is not orthogonal to the tissue [74]. While this strategy appeared promising in a few cases, its universal application was limited by directional OCT effects [75]. One strategy that avoids the errors in segmentation arising from typical drusen, RPD, the directional impact of non-orthogonal light was developed by Zhang et al. [76] (Fig. 6). This strategy used OAC-OCT imaging to segment the inner boundary of the outer plexiform layer (OPL) and measure the distance between this layer and RPE. The thickness within these segmentation boundaries served as an outer retinal layer (ORL) thickness measurement and a surrogate for photoreceptor thickness, and this strategy could be used in every stage of nonexudative AMD and would not be impacted by RPE elevations or the presence of RPD. This algorithm was used successfully to measure the outer retinal layer thickness around GA, and the ORL thickness measurement around GA at baseline was found to correlate with the future growth of GA as shown previously [70]. Further studies are underway at different stages of AMD to determine the usefulness of this ORL thickness measurement in predicting disease progression at different stages of AMD.

 

Fig. 6. Example of outer retinal layer (ORL) segmentation in regions around geographic atrophy (GA). A: Red dashed lines on the B-scan showing the inner boundary segmentation of the outer plexiform layer (OPL) and RPE. The yellow dashed lines show the boundaries of the slab located 64 µm to 400 µm beneath BM that was used to visualize GA based on the extent of the choroidal hypertransmission defect. B: Manual outline of GA derived from the en face sub-RPE slab generated yellow dashed lines in A. C: ORL thickness map with the GA outlined in red with the subregions around the GA margin outlined. R1 is represented by the green line that is 300 µm (1 degree) from the GA margin, and R2 is represented by the white line that is another 300 µm from the green line or 2 degrees from the margin of GA. (Figure reprinted with permission from Zhang et al. [76] in American Journal of Ophthalmology)

Download Full Size | PDF

8. Identification and quantitation of calcified drusen

Calcified drusen, an important risk factor for predicting disease progression in AMD, also cause obvious choroidal hypoTDs on en face OCT imaging using subRPE slabs [67,77]. However, up until recently, there has not been an algorithm to identify and quantitate calcified drusen. On OCT, calcified drusen are characterized by increased heterogeneous internal reflectivity compared with traditional soft drusen and by the presence of choroidal hypoTDs beneath the calcified drusen similar to the hypoTDs observed with hyperpigmentation, therefore OAC-based OCT imaging is ideally suited for the detection of calcified lesions [78]. Lu et al. [78] developed an OAC-based OCT algorithm that utilizes the OAC elevation map, as previously described, but they found that the anterior border of the RPE overlying these calcified drusen can be incorrectly segmented due to the high internal reflectivity of the drusen. The authors devised a novel correction strategy to estimate the correct anterior segmentation of these drusen and they then created an en face mean OAC projection map of these lesions. Their algorithm then identified the foci that were associated with calcified drusen by using a sub-choroid slab to select the hypoTDs that result from these calcified drusen (Fig. 7). By finding the overlap between the en face corrected OAC elevation map with the hypoTDs detected on the thresholded sub-choroid map, they were able to identify the calcified drusen that compared well to manual grading of the same scans. This novel algorithm can now be used to help identify the burden of calcified drusen in eyes with AMD and help determine the overall risk for disease progression in these eyes.

 

Fig. 7. The workflow for the automatic detection of calcified drusen from SS-OCT volume scan involves (A) creation of optical attenuation coefficient (OAC) B-scans with the segmentation lines positioned six pixels below the retinal pigment epithelium (RPE, the blue line) to two pixels above Bruch’s membrane (BM, the orange line) (B) En face map of drusen displaying the distances between BM and the RPE. (C) En face maximum mean projection of the OAC image from the slab defined by the two lines shown in (A). (D, E) Representative OAC B-scans with their locations marked by dashed red lines in (B, C) overlaid with the segmentation lines of six pixels below RPE (the blue line) to two pixels above BM (the orange line), showing small deviations in the RPE segmentations leading to false anterior segmentation of the calcified drusen due to the high internal OAC values (the green arrow). To provide volume and area measurements of the calcified drusen, the anterior boundaries of the calcified drusen are adjusted in the algorithm by applying a correction method described in the text [78]. (F) The first binary mask derived from (C) with some erroneously identified calcified drusen due to imperfect segmentation of the RPE at some locations. (G) OCT B-scans with the segmentation lines (the yellow lines) located within the sclera region to define the slab thickness of 40 µm, where the first line is determined by shifting the BM segmentation line by the maximum choroidal thickness in the volume. This sub-choroidal slab in the scleral region identified the hypoTDs resulting from the calcified drusen. (H) En face image of the sub-choroidal OCT slab is defined the two lines shown in (G). (I) The second binary mask derived from (H). (J) The final binary mask resulting from the overlap between (F) and (I) to indicate the regions of true calcified drusen. (Figure reprinted with permission from Lu et al. [78] in Biomedical Optics Express)

Download Full Size | PDF

9. Limitations of current OAC algorithms and future directions

While the theoretical framework of the OAC method provides a foundation for using optical attenuation (i.e. optical scattering property) as a means to distinguish between different physiological layers, the depth-resolved OAC [57] is known to have inaccurate estimations due to practical imaging conditions that can violate its assumptions under which the theoretical formulation is based. For example, one of the assumptions is that all light is attenuated at the bottom of the OCT image, which may not be achieved in practical retinal imaging despite the extended ranging distance of recent OCT systems. In this effort, Liu et al [79] provided a solution to estimate the OAC values more accurately where the original depth-resolved OAC formulation [57] is combined with a seeded optical attenuation coefficient for the deepest layer in the OCT scan estimated by curve fitting to the Beer Lambert law. Such treatment would relax the assumption that all the light must be attenuated at the bottom of the OCT image. While the accurate estimation of OAC values is an active area of research, it remains to be confirmed and validated whether it would affect the segmentation of macular layers and whether the OAC values are clinically meaningful in the diagnosis and treatment of ocular diseases.

Similarly, the marked attenuation that results from overlying structures found in AMD such as vitreous floaters, hemorrhage, fibrosis, hyperpigmentation, calcified drusen, and other RPE detachments that significantly attenuate the light penetrating the retina and choroid cannot be corrected using OAC enhancement. Another limitation is that unlike the fully automated algorithms that use optical attenuation correction for choroidal imaging and the OAC elevation algorithms, the remaining algorithms that include the detection of hyperTDs, calcified drusen, and the hyperpigmentation burden are semi-automated and still require some minor manual editing. Future directions will attempt to fully automate these existing algorithms.

In addition, all these algorithms have been developed for SS-OCTA volumetric scans obtained with a scanning speed of 100kHz, and future directions will include modifications for use with dense, volumetric SD-OCTA scans, and SS-OCTA scans obtained with faster scanning speeds. Furthermore, because reproducible optical attenuation coefficient calculations depend on the confocal function [80–82] and wavelength of the imaging system, the applicability of the algorithms across instruments provided by different commercial vendors remains to be determined.

Additional future directions include the use of OAC-enhanced OCT imaging to develop algorithms for better visualization of reticular pseudodrusen, also known as subretinal drusenoid deposits, [83,84] and the use of OAC enhancement for the detection of ocular blood flow. In addition, the current OAC elevation algorithm that is capable of detecting BLDs might be useful for following type 1 MNV to determine if the change in the RPE elevation corresponding to the extent of neovascularization might serve as a predictor of near-term exudation.

10. Conclusions

The use of OACs in OCT imaging has greatly improved our ability to identify and quantitate choroidal thickness, choroidal vessels, choroidal hyperTDs, BLDs, hyperpigmentation, outer retinal thickness, and calcified drusen. By developing algorithms that combine OACs along with intensity-based information in dense OCT volume scans, we have been able to identify and quantitate important anatomic changes that facilitate the staging of AMD and help predict disease progression. These algorithms will help identify patients at increased risk for disease progression, help quantify risk factors to improve subject selection for clinical studies and help with the design of endpoints for studies to determine if novel therapies can slow or prevent disease progression.