Open Access
Add to BibSonomy
Abstract
An automated depth-resolved algorithm using optical attenuation coefficients (OACs) was developed to visualize, localize, and quantify hyperreflective foci (HRF) seen on OCT imaging that are associated with macular hyperpigmentation and represent an increased risk of disease progression in age related macular degeneration. To achieve this, we first transformed the OCT scans to linear representation, which were then contrasted by OACs. HRF were visualized and localized within the entire scan by differentiating HRF within the retina from HRF along the retinal pigment epithelium (RPE). The total pigment burden was quantified using the en face sum projection of an OAC slab between the inner limiting membrane (ILM) to Bruch’s membrane (BM). The manual total pigment burden measurements were also obtained by combining manual outlines of HRF in the B-scans with the total area of hypotransmission defects outlined on sub-RPE slabs, which was used as the reference to compare with those obtained from the automated algorithm. 6×6 mm swept-source OCT scans were collected from a total of 49 eyes from 42 patients with macular HRF. We demonstrate that the algorithm was able to automatically distinguish between HRF within the retina and HRF along the RPE. In 24 test eyes, the total pigment burden measurements by the automated algorithm were compared with measurements obtained from manual segmentations. A significant correlation was found between the total pigment area measurements from the automated and manual segmentations (P < 0.001). The proposed automated algorithm based on OACs should be useful in studying eye diseases involving HRF.
© 2022 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement
1. Introduction
Hyperreflective foci (HRF) detected by optical coherence tomography (OCT) imaging of eyes with age-related macular degeneration (AMD) result from the aggregation and migration of cells from the retinal pigment epithelium (RPE) into the retina [1]. These HRF appear as foci of hyperpigmentation on color fundus imaging (CFI) and fundus biomicroscopy, and when associated with medium-sized drusen, these pigmentary abnormalities are sufficient to categorize an eye as having intermediate AMD (iAMD) [2,3]. Moreover, these HRF are recognized as a risk factor for disease progression from iAMD to late AMD [4–9]. A recent OCT study found that focal pigment abnormalities on CFI not only correspond to intraretinal HRF, but also to HRF along the RPE [10]. Due to the high reflectivity of HRF in the retina and along the RPE, these lesions are associated with decreased light penetration into the choroid and cause OCT choroidal hypo-transmission defects (hypoTDs), which appear as dark foci on en face sub-RPE choroidal slabs [10]. In light of these observations, we will use the terms foci of hyperpigmentation and pigment burden to refer to the HRF on OCT.
While the presence of HRF has been recognized as a risk factor for disease progression in AMD, there have been limited attempts to quantify these lesions using OCT imaging [7,10]. In these previous reports [7,11], retinal OCT B-scans and en face images were used to identify HRF in the retina, but foci of hyperreflectivity along or in close proximity to the RPE were not included. This oversight significantly limits the utility of these approaches to quantify the total burden of hyperpigmentation, which, we now know, includes foci of intraretinal hyperpigmentation and hyper-pigmentation along the RPE [10]. Currently, we are unaware of any automated algorithms that can visualize and quantify this total retinal pigment burden in the macula.
In order to develop a rapid and reliable algorithm that could quantify the total pigment burden in the macula, we developed a novel approach for identifying both the HRF in the retina and the foci of hyperpigmentation along the RPE. These lesions were localized and quantified by identifying the foci of hyperpigmentation within OCT slabs that included only the retina (above RPE) and foci of hyperpigmentation within slabs that included only the RPE complex, while the total pigment burden was quantified using a slab that included both the retina and RPE. Reference manual total pigment burden measurements were obtained by combining manual outlines of HRF within the B-scans and the total manual outlines of hypoTDs visualized on en face sub-RPE slabs that used segmentation boundaries extending from 64 to 400 µm below Bruch’s membrane (BM).
Previously, we pioneered the use of these sub-RPE slabs to visualize and quantify areas of choroidal hyper-transmission defects (hyperTDs) associated with the formation and growth of geographic atrophy (GA) [12–16]. While using these slabs to characterize GA, we came to appreciate that the same sub-RPE en face slabs could be used to identify areas of choroidal hypoTDs in iAMD, which appeared as dark regions on the en face OCT images [10,17]. While hypoTDs are often associated with hyperpigmentation, there are other pathological features such as calcified drusen, vitelliform material, and large retinal pigment epithelial detachments (PEDs) that may also lead to choroidal hypoTDs in eyes with iAMD [17–19]. These non-pigmentary lesions can be easily identified and excluded from the manual annotations by reviewing the corresponding B-scan through a particular region [10,17]. This strategy of using OCT en face imaging in conjunction with B-scans is currently used clinically to identify and localize the pathologies associated with both choroidal hyperTDs and hypoTDs. An automated algorithm for the segmentation of these pigmentary lesions would reduce the labor required to manually identify the pathologies associated with hypoTDs.
In this report, we utilize a strategy that converts OCT images into images that are contrasted by optical attenuation coefficients (OAC), a technique which has been shown to be useful in the automated segmentation and evaluation of GA [20,21], as well as choroidal thickness and choroidal vascularity [22,23]. OAC is defined in terms of the loss of OCT signal with depth caused by tissue absorption and scattering [24,25]. It is one of the physical parameters that measures the optical properties of a tissue [26] and can be directly extracted from OCT scans. OAC values enhance the contrast between high and low scattering regions in comparison to the standard OCT intensity and can be useful in localizing and quantifying the total pigment burden in the macula, since the HRF are characterized by strong optical scattering properties. Herewith, we describe a strategy using the OAC information from the original OCT images to develop an automated algorithm to visualize, localize, and quantify HRF in the retina and along the RPE by using the OACs within specific slabs obtained from the entire OCT scanning volume.
2. Method
2.1. Imaging acquisition
Patients at the Bascom Palmer Eye institute were enrolled in a prospective SS-OCT imaging study approved by the institutional review board of the University of Miami, Miller School of Medicine. Informed consent was obtained from each subject before imaging. The study was performed in accordance with the tenets of the Declaration of Helsinki and complied with the Health Insurance Portability and Accountability Act of 1996. For each enrolled patient, they underwent SS-OCT imaging using a 6×6 mm macular scan pattern (PLEX Elite 9000; Carl Zeiss Meditec, Inc, Dublin, CA, USA). The SS-OCT instrument has a scanning rate of 100,000 A-scans per second and a swept source laser with a central wavelength of 1,050 nm. Each 6×6 mm scan consisted of 500 B-scans, repeated twice at each B-scan location, and each B-scan consisted of 500 A-scans. Scans with a signal strength less than 7 or with obvious motion artifacts were excluded. Scans were then retrospectively reviewed for the presence of HRF consistent with increased pigmentary deposits within the retina and along the RPE.
2.2. OAC calculation from OCT scan
Multiple models and methods have been developed to evaluate OAC from OCT scans [24,26–29]. In this work, SS-OCT scans were converted into OAC maps using the method described by Vermeer et al [30]:
where A is the OCT signal intensity in linear space and i denotes the pixel’s position on A-Scan. Δ is the pixel size, and µ is the OAC which has a unit of mm-1. Regions of HRF have higher OACs due to greater light attenuation than the surrounding tissue. Representative OCT B-scans and corresponding OAC images are shown in Fig. 1, where the highly scattering RPE and migrated pigment are more strongly contrasted compared to the original OCT B-scans.
Fig. 1. Representative OCT B-scans of eyes with (A) drusen, (B) intra-retinal pigment migration (blue arrow) and (C) hyperreflective foci (HRF) along the retina pigment epithelium (RPE, red arrow), respectively. Corresponding optical attenuation coefficient (OAC) B-scans with (D) drusen, (E) intra-retinal pigment migration (blue arrow) and (F) HRF along the RPE (red arrow). Scale bar: 1 mm.
2.3. Visualization of HRF in 3-dimensional OAC volume
The distinct contrast of HRF in the OAC transformed OCT scans facilitates their automatic visualization and quantification. To do so, the segmentations of inner limiting membrane (ILM), nerve fiber layer (NFL), RPE and Bruch’s membrane (BM) are required. While there are numerous strategies developed to achieve the segmentations including machine learning algorithms [31–39], in this work, the segmentations of ILM, NFL, RPE and BM were obtained using the manufacturer’s automated software from OCT scan (Fig. 2(A)). The OAC data of the entire scan was then flattened at the position of BM for each B-scan (Fig. 2(C)). The 99.9 percentile of the intensities of all the OAC signals in the slab between ILM and NFL was set as the threshold for the background. Any signal in the OAC dataset with intensity smaller than the threshold was set to zero. Three channels were used to color code the information to help facilitate the visualization of the location of the HRF within the retina: (1) A mask of the slab from ILM to 40 µm above RPE was generated to identify intra-retinal HRF. The OAC values within this masked region were color-coded from black to blue, so that pigment migration is highlighted in blue (Figs. 2(D)&E), (2) A mask of the slab from 40 µm above RPE to 10 µm below RPE was generated to identify HRF along the RPE. The OAC values within the masked region were color-coded from black to red, so that pigment deposits were highlighted with a red color (Figs. 2(F)&G). (3) A slab below RPE with a thickness of 5 pixels (10 µm) is generated to demonstrate the RPE elevation in the region. The pixels in each A-scan were assigned as the distance from the RPE to BM at the location and color coded from grey to white, in which the regions with drusen were highlighted in white (Figs. 2(H)&I). All the parameters were set empirically to best demonstrate the HRF and their surrounding tissues.
Fig. 2. Color-coded visualization of hyperreflective foci (HRF) using optical attenuation coefficients (OACs). (A) OCT B-scan overlaid with automated segmentations of the internal limiting membrane (ILM), retinal pigment epithelium (RPE) and BM (Bruch’s membrane, yellow lines) from the manufacturer’s software. (B) OACs of the corresponding B-scan with overlaid segmentations of ILM, RPE and BM. (C) Flattened OAC B-scan based on BM with segmentations of ILM, RPE and BM. (D, E) A mask is generated for intra-retinal region between ILM to 40 µm above RPE illustrated as blue lines in D. RPE segmentation is illustrated as yellow lines. OAC signals inside the masked region are color-coded from black to blue, where intra-retinal pigment migration is highlighted in blue with a blue arrow (E). (F, G) A mask is generated for the RPE-involved region from 40 µm above RPE to 10 µm below RPE illustrated as red lines in F. OAC signals inside the masked region are color-coded from black to red, where increased pigmentation along RPE is highlighted in red with a red arrow (G). (H, I) A mask is generated as a 10 µm thick slab from 10 to 20 µm below RPE illustrated as white lines in H. RPE elevation is color-coded from grey to white, where regions with drusen are highlighted in white (I). (J) Combined images showing HRF in the retina (blue) and along RPE (red) with elevation information (greyscale). Blue arrow: HRF in the retina; red arrow: HRF along the RPE. Scale bar: 1 mm.
2.4. Quantification of HRF from en face images
The intra-retinal HRF from each OCT B-scan (Figs. 3(A)-(C)) and the areas with choroidal hypotransmission defects on OCT sub-RPE slabs were manually outlined by two graders (RL and JL) (Figs. 3(D)-(F)). The grading on B-scans were firstly projected to an en face binarized image (Fig. 3(C)) by marking all the pixels with the corresponding labeled A-line positions to 1 and then combined with the grading on sub-RPE images to generate the en face manual labels of the total pigment burden (Fig. 3(G)). The grading on B-scans is necessary because not all the intra-retinal HRF have hypo-transmission defects on choroid, therefore, these pigments might be missed if only sub-RPE images are graded. For automated quantification of the total retinal pigment burden using OAC imaging, an en face strategy was used. Similar to other en face strategies previously used for GA segmentation and area measurement [20,40], an en face sum projection of OAC signals in a slab from the ILM to BM was generated (Fig. 3(H), I). To segment the total pigment burden, we applied an adaptive thresholding strategy with a local window of 700 µm in diameter (Figs. 3(J), K). A MatLab built-in function was applied to calculate the locally adaptive threshold for each pixel using the local mean intensity around the neighborhood of the pixel. The area of the total pigment burden was defined within a 5 mm circle centered on the fovea, and this area was used to compare the measurements obtained from manual segmentation and the automated algorithm.
Fig. 3. Manual quantification of the total pigment burden obtained using OCT B-scans and en face images, and automated quantification of the total pigment burden obtained using optical attenuation coefficient (OAC) en face images. (A, B) The intra-retinal HRF were manually outlined on OCT B-scans throughout the entire volume (blue). (C) En face images (binarized) of intra-retinal HRF were generated from the manually outlined B-scans. The blue dashed line indicates the location of the corresponding B-scan in A and B. (D, E) The regions with hypo-transmission defects were manually outlined on the OCT slabs from beneath the retinal pigment epithelium (sub-RPE slabs; red). (F) En face image showing the binarized regions with the hypo-transmission defects. (G) Combined binarized en face map of the total pigment burden from the manual outlines. (H) OAC B-scan overlaid with segmentation boundaries of the ILM and BM (white dashed lines). (I) En face OAC image with a blue dashed line indicating the location of the corresponding B-scan. (J) En face OAC image overlaid with automatically segmented HRF (yellow). (K) Binarized image of total pigment burden detected automatically using the OACs. HRF: hyper-reflective foci. Scale bar: 1 mm.
2.5. Statistical analysis
Pearson correlation and Bland-Altman analysis were utilized to compare the area measurements of the total pigment burden from the manual segmentations and the automated algorithm. Statistical analysis was carried out using MATLAB R2018b and IBM SPSS V25 (Armonk, NY, USA), and plots were generated using GraphPad Prism (GraphPad Software, San Diego, CA, USA).
3. Results
A total of 49 eyes with nonexudative AMD from 42 patients (74.6 ± 6.9 years, 61.2% women) with intra-retinal HRF and increased reflectivity along the RPE were recruited. In these eyes hyper-reflective lesions were visualized using the automated algorithm developed in this study, intra-retinal HRF were distinguished from hyper-reflective lesions along the RPE, and the position of these HRF were localized relative to the position of drusen.
Figure 4 provides three examples of the visualization of HRF in eyes. Figures 4(A), (D), (G) show an eye of 84-year-old women with intra-retinal pigment aggregation and migration. HRF were observed both in the retina (blue) and along the RPE (red) within the macula. No drusen were present. Figures 4(B), (E), (H) show an eye of a 74-year-old man with less amount of intra-retinal pigment migration, but dense pigment deposits along the RPE. The deposits of hyperpigmentation along the RPE were associated with drusen. Figures 4(C), (F), (I) show an eye of 75-year-old women with no obvious intra-retinal HRF, but significant central pigmentary lesions along the RPE. In contrast to the patient showed in Fig. 4(B), the pigmentary lesions were centralized at the fovea.
Fig. 4. Visualization of HRF in the retina and along RPE and their relationship with surrounding drusen. (A-I) Side views (A-C), top view (D-F), and selected OCT B-scan views (G-I, with the locations of the B-scans is indicated by the yellow dashed lines in (D-F) with HRF 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 HRF and hyperpigmentation along the RPE. (B, E, H) An eye with drusen and HRF along the RPE (highlighted in red) that appear as foci of hypereflective material on the B-scan (H). (C, F, I) An eye with a central area of hyperreflective material along the RPE (highlighted in red). Scale bar: 1 mm.
The total pigment burden area measurements were automatically generated from the en face OAC images and validated against manual segmentations. Out of the 49 recruited eyes, the hyper-reflective lesions of a test set of 24 eyes from 20 patients (74.2 ± 6.4 years, 70.8%women) were manually segmented by identifying intra-retinal lesions on OCT B-scans and choroidal hypo-transmission defects on OCT sub-RPE slabs extending 64 to 400 µm below BM. A consensus was reached between the two graders (RL and JL) on all the 24 scans. Note that this set of scans with manual annotations was not used to develop the OAC algorithm or in optimizing the thresholding parameters. A significant correlation was found in the area measurements of the total pigment burden between the automated and manual segmentations (P < 0.001, Fig. 5(A)). Bland-Altman analysis showed that there was an averaged bias of 0.032 mm2 between manual and automated segmentations (Fig. 5(B)). The areas of pigment segmented from the OAC algorithm tended to be smaller than the manually segmented pigment from OCT scans. Overall, the automated algorithm had a good agreement with human graders.
Fig. 5. Comparison of the area of total pigment burden measured manually and automatically. (A) Area of total pigment burden measured using the OAC algorithm vs. the area of total pigment measured manually. (B) Bland-Altman analysis of the data shown in A.
Figures 6 and 7 show two representative eyes with HRF along the RPE (Fig. 6) and within the retina (Fig. 7). There was good agreement between the segmentations from the automated algorithm and those from the human graders (Figs. 6&7 D&H). The OAC B-scans show enhanced contrast of the hyperpigmentation that appear slightly smaller compared with the original OCT B-scans (Figs. 6&7 B&F), this might explain why the segmented regions from OAC images are generally smaller than the manually segmented regions from the OCT images. We also observed improved contrast of pigmentary lesions on the en face images using OACs (Figs. 6&7 A&E), and this might explain why some small lesions were picked up by the algorithm which were not marked by the graders.
Fig. 6. An example of manual and automated segmentations of the total pigment burden in an eye with hyper-reflective foci along the RPE. (A) En face OCT image of the sub-RPE slab. (B) Representative OCT B-scan (location indicated by yellow line). (C) Manual segmentation (red) of total pigment burden overlaid on the slab from beneath the retinal pigment epithelium (sub-RPE slab). The manual segmentation was generated combining the outlines from sub-RPE OCT image and B-scans. (D) Binarized total pigment burden from the manual segmentation. (E) En face OAC images of a slab from the ILM to BM. (F) Representative OAC B-scan (location indicated by yellow line). (G) Automated segmentation of the total pigment burden (yellow) overlaid on the OAC image. (H) Binarized total pigment burden from automated algorithm. Scale bar: 1 mm.
Fig. 7. An example of manual and automated segmentations of the total pigment burden in an eye with intra-retinal hyper-reflective foci. (A) En face OCT image of the sub-RPE slab. (B) Representative OCT B-scan (location indicated by yellow lines). (C) Manual segmentation (red) of total pigment burden overlaid on the slab from beneath the retinal pigment epithelium (sub-RPE slab). The manual segmentation was generated combining the outlines from sub-RPE OCT image and B-scans. (D) Binarized total pigment burden from the manual segmentation. (E) En face OAC images of a slab from the ILM to BM. (F) Representative OAC B-scan (location indicated by yellow lines). (G) Automated segmentation of the total pigment burden (yellow) overlaid on the OAC image. (H) Binarized total pigment burden from automated algorithm. Scale bar: 1 mm.
We observed that two out of the 24 tested eyes showed larger disagreements between manual and automated segmentations (See Fig. 5). Figure 8 shows the case where area of total pigment burden segmented from manual segmentation (0.08 mm2) is much smaller than that from the automated algorithm (0.16 mm2). Two lesions were identified (blue & yellow arrows) that caused most of the difference. The OCT and OAC B-scan of the lesion shown via the blue arrow (Figs. 8 C&G) demonstrated that there is a thickened RPE on the top of the drusen. Since the lesion does not cast a prominent hypoTD in the choroid, it was missed by the manual grading criteria (solid blue square). However, the algorithm using OACs was able to identify the lesion due to the higher attenuation of light at the thickened RPE (dashed blue square). The OCT and OAC B-scans of the lesion shown via the yellow arrow (Figs. 8(D)&H) demonstrated that there is a highly-elevated RPE that is neglected by the manual grading criteria because of the mild hypoTD in the choroid. The algorithm was again able to identify such enhanced light attenuation (yellow dashed square).
Fig. 8. An example showing the area of the total pigment burden segmented by the algorithm being larger than the area from the manual segmentation. (A) En face OCT image of a slab from beneath the retinal pigment epithelium (sub-RPE slab). (B) Binarized total pigment burden from the manual segmentation. (C, D) OCT B-scans (two representative locations indicated by blue and yellow lines). (E) En face OAC image of the slab from the ILM to BM (F) Binarized total pigment burden from the automated segmentation. (G, H) OAC B-scans (two representative locations indicated by blue and yellow lines). Solid & dashed squares showed the manual and automated segmentations respectively in the two representative B-scans. The blue & yellow arrows showed the difference between the manual and automated segmentations, respectively, with the automated algorithm identifying poorly defined hyperreflective material along the RPE. Scale bar: 1 mm.
Figure 9 shows the other case where area of total pigment burden segmented from manual segmentation (0.24 mm2) is much larger than that from the automated algorithm (0.06 mm2). Two major differences were identified (blue & yellow arrows) comparing the en face binarized label of pigments (Figs. 9(B)&F). The OCT and OAC B-scans of the lesion shown via the blue arrow (Figs. 9(C)&G) demonstrated a thickened RPE on the top of the drusen and a mild aggregation of pigments in the retina. The lesions casted hypoTD in the choroid and intraretina pigmentation was identified by the manual grading criteria (blue solid squares). However, the OAC value showed a relatively low contrast with the surrounding regions so that the algorithm fails to pick it up as regions with pigment clumping. The OCT and OAC B-scan of the lesion shown via the yellow arrow (Figs. 9(D)&H) demonstrated multiple pigment migrations in the retina. Since the contrast was enhanced by OAC, the algorithm was able to successfully identify the two separated lesions with smaller area instead of a large, combined lesion from the OCT B-scan (segmented regions highlighted by yellow solid and dashed squares). The similar bias of area measurement of pigments was also observed in the Bland-Altman analysis in previous result (Fig. 5(B)).
Fig. 9. An example showing the area of the total pigment burden segmented by algorithm being smaller than the area from the manual segmentation. (A) En face OCT image of a slab from beneath the retinal pigment epithelium (sub-RPE slab). (B) Binarized total pigment burden from the manual segmentation. (C, D) OCT B-scans (two representative locations indicated by blue and yellow lines). (E) En face OAC image of the slab from the ILM to BM (F) Binarized total pigment burden from the automated segmentation. (G, H) OAC B-scans (two representative locations indicated by blue and yellow lines). Solid & dashed squares showed the manual and automated segmentations respectively in the two representative B-scans. The blue & yellow arrows showed the difference between the manual and automated segmentations, respectively, with the automated algorithm missing lesions when comparing C and G, and identifying a smaller area of hyperpigmentation compared with the manual segmentations in another lesion, which is most obvious when comparing D and H. Scale bar: 1 mm.
4. Discussion
Both the OAC B-scans and en face maps have demonstrated their importance in complementing traditional OCT images so that the physical properties of tissues can be enhanced and the more prominent features can be used to more easily develop automated algorithms [26,41]. Previously, OACs have been applied to improve the automated segmentation of GA and the choroid [26,41]. For the purpose of visualizing and automatically quantifying HRF, we sought to use OACs to identify regions where the light attenuation is the greatest at different depths between the ILM and just beneath the RPE. This strategy provided us with depth-resolved information and allowed us to distinguish intraretinal HRF from foci of increased pigmentation along the RPE since both appear as hyperpigmentation on color fundus imaging and as hypotransmission defects on sub-RPE OCT slabs [10]. Furthermore, since the OAC is calculated as a relative ratio of attenuation along the light path, the variance in laser energy and signal strength between scans would have a minimal effect on the measurements [25,26].
There have been previous attempts to develop automated algorithms to segment the HRF in the retina from OCT B-scans [11,42–\44]. One group developed a complicated algorithm combining regions of interest generated by using morphological reconstruction, and the HRF were estimated using a component tree method to quantify the HRF from B-scans on spectral domain OCT (SD-OCT) [43]. However, the noise from retinal vessels and disrupted layers made the automated segmentation extremely challenging. In addition, several deep convolutional neural networks have been developed to segment the HRF in the retina from OCT B-scans [11,42,44]. The major limitation of these deep learning methods is the laborious manual annotations on cross-sectional images, making the performance of the model highly dependent on the small dataset in the study. So far, no automated algorithms are available in the clinical settings that can satisfactorily visualize and quantify the intraretinal HRF and the increased pigmentation along the RPE, to provide an assessment of the total retinal pigment burden.
The ability to provide depth-resolved visualization of HRF should be useful in distinguishing intraretinal pigment migration and hyperpigmentation along the RPE to determine if either or both serve as risk factors for disease progression and to provide a convenient risk-assessment for disease progression in eyes with iAMD. Of note, it has been reported that intra-retinal RPE migration mainly occurred above areas of drusen (73.3% of eyes) [45]. As demonstrated in Fig. 4, our proposed technique provided a visual representation of the entire scanned volume resulting in the opportunity to directly identify and quantitate where the hyperpigmentation is located and distributed within the macula. The proposed method using OACs to evaluate reflectivity along RPE should be useful to distinguish the non-pigmentary lesions such as calcified drusen, vitelliform material, and large retinal PEDs in eyes with iAMD because such lesions may also lead to choroidal hypoTDs in OCT scans but not necessarily high OACs. It is worth exploring whether OACs can work as a biomarker to separate pigmentary lesions from non-pigmentary lesions in the future study.
The quantification of HRF from en face OAC images has several advantages. It is consistent with several other clinical practices using en face images in AMD studies such as drusen mapping, the identification and segmentation of GA, and the analysis of choriocapillaris flow deficits. By using only one en face image for each scan, it is faster to compute compared with previous methods using hundreds of OCT B-scans for a single volumetric scan [11,42–\44]. We validated our strategy by showing that the measurements of total pigment burden using OACs were highly correlated with those from manual segmentations. While the results illustrated the differences between automated and manual segmentations in some complicated cases, these differences were mainly the results of how the foci of pigmentation appear on the OAC images versus traditional OCT image, with the OAC foci appearing slightly smaller. This can be further improved by optimizing the parameters using a larger and more various training dataset and by considering to combine automated strategies on both OAC and OCT images. Thus, by using the OAC strategy and an automated thresholding method, we were able to largely eliminate the laborious measurement of areas with increased pigmentation on OCT B-scans and areas of hypotransmission defects on sub-RPE en face OCT images.
There are several limitations in this study. First, the current work only validated the quantification of total pigment burden from en face OAC images. It will be interesting to explore if the technique can further provide quantification of HRF in the retina and along the RPE separately. Previous studies have reported using OCT B-scans that the HRF would most frequently migrate from the RPE into the outer nuclear layer (66.7% of eyes) and less frequently into more anterior retinal layers [45]. With the en face OAC images of selected slabs, it will be possible to easily study the dynamics of pigment migration over time without the tedious review on OCT B-scans at numerous visits. In addition, while the proposed method focusses on the HRF along and above the RPE since these lesions appear to be associated with the greatest risk of disease progression in AMD, it should also be possible to detect HRF in the choroid as previously reported from OCT B-scans [46]. Another limitation is that the scanning protocol used in this study was limited to 6 × 6 mm macular scans of SS-OCT only. Based on the properties of OACs, it should be possible to apply the same technique to other scanning protocols with different fields-of-view, as well as other OCT imaging modalities such as SD-OCT. Lastly, a limited number of scans have been tested and used for validation. Further testing with a larger dataset of eyes that contain various pathologies is warranted to provide additional validation of the technique and demonstrate its potential broader applications in clinical settings.
5. Conclusion
Overall, an automated algorithm based on OACs was able to visualize, localize, and quantify areas of increased pigmentation in the retina and along the RPE, and the automated en face measurements of the total pigment burden were successfully validated against manual measurements. This algorithm should be useful for assessing the overall risk of disease progression in eyes with AMD, monitoring disease progression, and studying the impact of therapies in eyes with AMDs.
Funding
Carl Zeiss Meditec inc; Salah Foundation; Research to Prevent Blindness; National Eye Institute (P30EY014801, R01EY028753).
Disclosures
Dr. Gregori, Dr. Rosenfeld and Dr. Wang received research support from Carl Zeiss Meditec, Inc. Dr. Gregori and the University of Miami co-own a patent that is licensed to Carl Zeiss Meditec, Inc. Dr. Gregori and Dr. Rosenfeld received support the National Eye Institute Center Core Grant (P30EY014801) and Research to Prevent Blindness (unrestricted Grant) to the Department of Ophthalmology, University of Miami Miller School of Medicine.
Dr. Rosenfeld also received research funding from Gyroscope Therapeutics and Stealth BioTherapeutics. He is also a consultant for Boehringer-Ingelheim, Carl Zeiss Meditec, Chengdu Kanghong Biotech, InflammX/Ocunexus Therapeutics, Ocudyne, Regeneron Pharmaceuticals, and Unity Biotechnology. He also has equity interest in Apellis, Valitor, Verana Health, and Ocudyne. Dr. Wang discloses intellectual property owned by the Oregon Health and Science University and the University of Washington. Dr. Wang also receives research support from Moptim Inc, and Colgate Palmolive Company. He is a consultant to Carl Zeiss Meditec. Dr. Zhou, Mr. Liu, Dr. Laiginhas, Dr. Zhang, Mr. Cheng, Mrs. Zhang, Dr. Shi and Dr. Shen have no disclosures.
Data availability
Data underlying the results presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.
References
1. D. F. Cao, B. Leong, J. D. Messinger, D. Kar, T. Ach, L. A. Yannuzzi, K. B. Freund, and C. A. Curcio, “Hyperreflective Foci, Optical Coherence Tomography Progression Indicators in Age-Related Macular Degeneration, Include Transdifferentiated Retinal Pigment Epithelium,” Invest. Ophthalmol. Visual Sci. 62(10), 34 (2021). [CrossRef]
2. M. D. Davis, R. E. Gangnon, L. Y. Lee, L. D. Hubbard, B. E. K. Klein, R. Klein, F. L. Ferris, S. B. Bressler, R. C. Milton, and A.-R. E. D. S. Res, “The age-related eye disease study severity scale for age-related macular degeneration - AREDS report no. 17,” Arch Ophthalmol 123(11), 1484–1498 (2005). [CrossRef]
3. F. L. Ferris, C. P. Wilkinson, A. Bird, U. Chakravarthy, E. Chew, K. Csaky, S. R. Sadda, and B. I. M. Res, “Clinical Classification of Age-related Macular Degeneration,” Ophthalmology 120(4), 844–851 (2013). [CrossRef]
4. J. G. Christenbury, F. A. Folgar, R. V. O’Connell, S. J. Chiu, S. Farsiu, C. A. Toth, A.-r. E. d., and S. Ancill, “Progression of Intermediate Age-related Macular Degeneration with Proliferation and Inner Retinal Migration of Hyperreflective Foci,” Ophthalmology 120(5), 1038–1045 (2013). [CrossRef]
5. K. Sleiman, M. Veerappan, K. P. Winter, M. N. McCall, G. Yiu, S. Farsiu, E. Y. Chew, T. Clemons, C. A. Toth, and A.-R. E. D. S. Ancil, “Optical Coherence Tomography Predictors of Risk for Progression to Non-Neovascular Atrophic Age-Related Macular Degeneration,” Ophthalmology 124(12), 1764–1777 (2017). [CrossRef]
6. J. Q. Lei, S. Balasubramanian, N. S. Abdelfattah, M. G. Nittala, and S. R. Sadda, “Proposal of a simple optical coherence tomography-based scoring system for progression of age-related macular degeneration,” Graefe's Arch. Clin. Exp. Ophthalmol. 255(8), 1551–1558 (2017). [CrossRef]
7. M. Nassisi, W. Y. Fan, Y. Shi, J. Q. Lei, E. Borrelli, M. Ip, and S. R. Sadda, “Quantity of Intraretinal Hyperreflective Foci in Patients With Intermediate Age-Related Macular Degeneration Correlates With 1-Year Progression,” Invest. Ophthalmol. Visual Sci. 59(8), 3431–3439 (2018). [CrossRef]
8. M. Nassisi, J. Q. Lei, N. S. Abdelfattah, A. Karamat, S. Balasubramanian, W. Y. Fan, A. Uji, K. M. Marion, K. Baker, X. W. Huang, E. Morgenthien, and S. R. Sadda, “OCT Risk Factors for Development of Late Age-Related Macular Degeneration in the Fellow Eyes of Patients Enrolled in the HARBOR Study,” Ophthalmology 126(12), 1667–1674 (2019). [CrossRef]
9. C. A. Curcio, E. C. Zanzottera, T. Ach, C. Balaratnasingam, and K. B. Freund, “Activated Retinal Pigment Epithelium, an Optical Coherence Tomography Biomarker for Progression in Age-Related Macular Degeneration,” Invest. Ophthalmol. Visual Sci. 58(1), 211–226 (2017). [CrossRef]
10. R. Laiginhas, J. Liu, M. Shen, Y. Shi, O. Trivizki, N. K. Waheed, G. Gregori, and P. J. Rosenfeld, “Multimodal Imaging, OCT B-Scan Localization, and En Face OCT Detection of Macular Hyperpigmentation in Eyes with Intermediate Age-Related Macular Degeneration,” Ophthalmology Science 2(2), 100116 (2022). [CrossRef]
11. L. Varga, A. Kovacs, T. Grosz, G. Thury, F. Hadarits, R. Degi, and J. Dombi, “Automatic segmentation of hyperreflective foci in OCT images,” Computer Methods and Programs in Biomedicine 178, 91–103 (2019). [CrossRef]
12. K. B. Schaal, P. J. Rosenfeld, G. Gregori, Z. Yehoshua, and W. J. Feuer, “Anatomic Clinical Trial Endpoints for Nonexudative Age-Related Macular Degeneration,” Ophthalmology 123(5), 1060–1079 (2016). [CrossRef]
13. K. B. Schaal, G. Gregori, and P. J. Rosenfeld, “En Face Optical Coherence Tomography Imaging for the Detection of Nascent Geographic Atrophy,” Am. J. Ophthalmol. 174, 145–154 (2017). [CrossRef]
14. Y. Shi, Q. Zhang, H. Zhou, L. Wang, Z. Chu, X. Jiang, M. Shen, M. Thulliez, C. Lyu, W. Feuer, L. de Sisternes, M. K. Durbin, G. Gregori, R. K. Wang, and P. J. Rosenfeld, “Correlations Between Choriocapillaris and Choroidal Measurements and the Growth of Geographic Atrophy Using Swept Source OCTImaging,” Am. J. Ophthalmol. 224, 321–331 (2021). [CrossRef]
15. Y. Shi, J. Yang, W. Feuer, G. Gregori, and P. J. Rosenfeld, “Persistent Hypertransmission Defects on En Face OCT Imaging as a Stand-Alone Precursor for the Future Formation of Geographic Atrophy,” Ophthalmology Retina 5(12), 1214–1225 (2021). [CrossRef]
16. S. B. Velaga, M. G. Nittala, A. Hariri, and S. R. Sadda, “Correlation between Fundus Autofluorescence and En Face OCT Measurements of Geographic Atrophy,” Ophthalmology Retina (2022).
17. J. Liu, R. Laiginhas, M. Shen, Y. Shi, J. Li, O. Trivizki, N. K. Waheed, G. Gregori, and P. J. Rosenfeld, “Multimodal Imaging and En Face OCT Detection of Calcified Drusen in Eyes with Age-Related Macular Degeneration,” Ophthalmology Science 2(2), 100162 (2022). [CrossRef]
18. R. F. Spaide, K. Noble, A. Morgan, and K. B. Freund, “Vitelliform macular dystrophy,” Ophthalmology 113(8), 1392–1400.e4 (2006). [CrossRef]
19. H. Zhou, J. Lu, K. Chen, Y. Y. Shi, G. Gregori, P. J. Rosenfeld, and R. K. Wang, “Mitigating the effects of choroidal hyper- and hypo-transmission defects on choroidal vascularity index assessments using optical coherence tomography,” Quant Imaging Med Surg 12(5), 2932–2946 (2022). [CrossRef]
20. Z. Chu, L. Wang, X. Zhou, Y. Shi, Y. Cheng, R. Laiginhas, H. Zhou, M. Shen, Q. Zhang, L. de Sisternes, A. Y. Lee, G. Gregori, P. J. Rosenfeld, and R. K. Wang, “Automatic geographic atrophy segmentation using optical attenuation in OCT scans with deep learning,” Biomed. Opt. Express 13(3), 1328–1343 (2022). [CrossRef]
21. Z. Chu, Y. Shi, X. Zhou, L. Wang, H. Zhou, R. Laiginhas, Q. Zhang, Y. Cheng, M. Shen, L. de Sisternes, M. K. Durbin, W. Feuer, G. Gregori, P. J. Rosenfeld, and R. K. Wang, “Optical Coherence Tomography Measurements of the Retinal Pigment Epithelium to Bruch Membrane Thickness Around Geographic Atrophy Correlate With Growth,” Am. J. Ophthalmol. 236, 249–260 (2022). [CrossRef]
22. H. Zhou, Y. Dai, Y. Shi, J. F. Russell, C. Lyu, J. Noorikolouri, W. J. Feuer, Z. Chu, Q. Zhang, L. de Sisternes, M. K. Durbin, G. Gregori, P. J. Rosenfeld, and R. K. Wang, “Age-Related Changes in Choroidal Thickness and the Volume of Vessels and Stroma Using Swept-Source OCT and Fully Automated Algorithms,” Ophthalmology Retina 4(2), 204–215 (2020). [CrossRef]
23. H. Zhou, Y. N. Dai, G. Gregori, P. R. Rosenfeld, J. L. Duncan, D. M. Schwartz, and R. K. K. Wang, “Automated morphometric measurement of the retinal pigment epithelium complex and choriocapillaris using swept source OCT,” Biomed. Opt. Express 11(4), 1834–1850 (2020). [CrossRef]
24. P. Gong, M. Almasian, G. van Soest, D. de Bruin, T. van Leeuwen, D. Sampson, and D. Faber, “Parametric imaging of attenuation by optical coherence tomography: review of models, methods, and clinical translation,” J. Biomed. Opt. 25(04), 1 (2020). [CrossRef]
25. J. M. Schmitt, A. Knuttel, M. Yadlowsky, and M. A. Eckhaus, “Optical-Coherence Tomography of a Dense Tissue - Statistics of Attenuation and Backscattering,” Phys. Med. Biol. 39(10), 1705–1720 (1994). [CrossRef]
26. S. Chang and A. K. Bowden, “Review of methods and applications of attenuation coefficient measurements with optical coherence tomography,” J. Biomed. Opt. 24(09), 1 (2019). [CrossRef]
27. B. Ghafaryasl, K. Vermeer, J. Kalkman, T. Callewaert, J. de Boer, and L. van Vliet, “Attenuation coefficient estimation in Fourier-domain OCT of multi-layered phantoms,” Biomed. Opt. Express 12(5), 2744–2758 (2021). [CrossRef]
28. K. Li, W. Liang, Z. Yang, Y. Liang, and S. Wan, “Robust, accurate depth-resolved attenuation characterization in optical coherence tomography,” Biomed. Opt. Express 11(2), 672–687 (2020). [CrossRef]
29. M. Hohmann, B. Lengenfelder, D. Muhr, M. Späth, M. Hauptkorn, F. Klämpfl, and M. Schmidt, “Direct measurement of the scattering coefficient,” Biomed. Opt. Express 12(1), 320–335 (2021). [CrossRef]
30. K. A. Vermeer, J. Mo, J. J. A. Weda, H. G. Lemij, and J. F. de Boer, “Depth-resolved model-based reconstruction of attenuation coefficients in optical coherence tomography,” Biomed Opt Express 5(1), 322–337 (2014). [CrossRef]
31. D. Stromer, E. Moult, S. Chen, N. Waheed, A. Maier, and J. Fujimoto, “Correction propagation for user-assisted optical coherence tomography segmentation: general framework and application to Bruch’s membrane segmentation,” Biomed. Opt. Express 11(5), 2830–2848 (2020). [CrossRef]
32. J. Schottenhamml, E. Moult, S. Ploner, S. Chen, E. Novais, L. Husvogt, J. Duker, N. Waheed, J. Fujimoto, and A. Maier, “OCT-OCTA segmentation: combining structural and blood flow information to segment Bruch’s membrane,” Biomed. Opt. Express 12(1), 84–99 (2021). [CrossRef]
33. T. Heikka, B. Cense, and N. Jansonius, “Retinal layer thicknesses retrieved with different segmentation algorithms from optical coherence tomography scans acquired under different signal-to-noise ratio conditions,” Biomed. Opt. Express 11(12), 7079–7095 (2020). [CrossRef]
34. S. Borkovkina, A. Camino, W. Janpongsri, M. Sarunic, and Y. Jian, “Real-time retinal layer segmentation of OCT volumes with GPU accelerated inferencing using a compressed, low-latency neural network,” Biomed. Opt. Express 11(7), 3968–3984 (2020). [CrossRef]
35. D. Romo-Bucheli, P. Seeböck, J. Orlando, B. Gerendas, S. Waldstein, U. Schmidt-Erfurth, and H. Bogunović, “Reducing image variability across OCT devices with unsupervised unpaired learning for improved segmentation of retina,” Biomed. Opt. Express 11(1), 346–363 (2020). [CrossRef]
36. J. Li, P. Jin, J. Zhu, H. Zou, X. Xu, M. Tang, M. Zhou, Y. Gan, J. He, Y. Ling, and Y. Su, “Multi-scale GCN-assisted two-stage network for joint segmentation of retinal layers and discs in peripapillary OCT images,” Biomed. Opt. Express 12(4), 2204–2220 (2021). [CrossRef]
37. J. Wang, W. Li, Y. Chen, W. Fang, W. Kong, Y. He, and G. Shi, “Weakly supervised anomaly segmentation in retinal OCT images using an adversarial learning approach,” Biomed. Opt. Express 12(8), 4713–4729 (2021). [CrossRef]
38. A. Saxena, X. Yao, D. Wong, J. Chua, M. Ang, Q. Hoang, R. Agrawal, M. Girard, G. Cheung, L. Schmetterer, and B. Tan, “Framework for quantitative three-dimensional choroidal vasculature analysis using optical coherence tomography,” Biomed. Opt. Express 12(8), 4982–4996 (2021). [CrossRef]
39. S. Mukherjee, T. De Silva, P. Grisso, H. Wiley, D. Tiarnan, A. Thavikulwat, E. Chew, and C. Cukras, “Retinal layer segmentation in optical coherence tomography (OCT) using a 3D deep-convolutional regression network for patients with age-related macular degeneration,” Biomed. Opt. Express 13(6), 3195–3210 (2022). [CrossRef]
40. M. Thulliez, Q. Zhang, Y. Shi, H. Zhou, Z. Chu, L. de Sisternes, M. K. Durbin, W. Feuer, G. Gregori, R. K. Wang, and P. J. Rosenfeld, “Correlations between Choriocapillaris Flow Deficits around Geographic Atrophy and Enlargement Rates Based on Swept-Source OCT Imaging,” Ophthalmology Retina 3(6), 478–488 (2019). [CrossRef]
41. H. Zhou, Z. Chu, Q. Zhang, Y. Dai, G. Gregori, P. J. Rosenfeld, and R. K. Wang, “Attenuation correction assisted automatic segmentation for assessing choroidal thickness and vasculature with swept-source OCT,” Biomed. Opt. Express 9(12), 6067–6080 (2018). [CrossRef]
42. C. C. Yu, S. Xie, S. J. Niu, Z. X. Ji, W. Fan, S. T. Yuan, Q. H. Liu, and Q. Chen, “Hyper-reflective foci segmentation in SD-OCT retinal images with diabetic retinopathy using deep convolutional neural networks,” Med. Phys. 46(10), 4502–4519 (2019). [CrossRef]
43. I. P. Okuwobi, Z. Ji, W. Fan, S. T. Yuan, L. Bekalo, and Q. Chen, “Automated Quantification of Hyperreflective Foci in SD-OCT With Diabetic Retinopathy,” IEEE J. Biomed. Health Inform. 24(4), 1125–1136 (2020). [CrossRef]
44. S. Xie, I. P. Okuwobi, M. C. Li, Y. H. Zhang, S. T. Yuan, and Q. Chen, “Fast and Automated Hyperreflective Foci Segmentation Based on Image Enhancement and Improved 3D U-Net in SD-OCT Volumes with Diabetic Retinopathy,” Trans. Vis. Sci. Tech 9(2), 21 (2020). [CrossRef]
45. J. Ho, A. J. Witkin, J. Liu, Y. L. Chen, J. G. Fujimoto, J. S. Schuman, and J. S. Duker, “Documentation of Intraretinal Retinal Pigment Epithelium Migration via High-Speed Ultrahigh-Resolution Optical Coherence Tomography,” Ophthalmology 118(4), 687–693 (2011). [CrossRef]
46. S. Fragiotta, S. Abdolrahimzadeh, R. Dolz-Marco, Y. Sakurada, O. Gal-Or, and G. Scuderi, “Significance of Hyperreflective Foci as an Optical Coherence Tomography Biomarker in Retinal Diseases: Characterization and Clinical Implications,” J. Ophthalmology 2021, 1–10 (2021). [CrossRef]
