Bruch's membrane segmentation in oct volume

Abstract

To improve retina layer segmentation in optical coherence tomography (OCT) data.SOLUTION: OCT data is enhanced on the basis of mixture of the OCT data and OCT angiography (OCTA) data. Contrast in the OCT data is enhanced in a region where the OCT data and the OCTA data are not similar to each other and is reduced in a region where the OCT data and the OCTA data are similar to each other. A target retina layer in the OCT data is segmented on the basis of the enhanced data. An error in the segmentation of the target retina layer in the OCT data is checked by using two en face images of the OCTA data including the target retina layer. The identified error is replaced with an approximate value based on positions of the highest retina layer and the lowest retina layer in one of the en face images.SELECTED DRAWING: Figure 16

Description

[Technical Field]

[0001] The present invention relates generally to automatic retinal layer segmentation in OCT data, and more particularly to segmenting retinal layers that lack definition in OCT data for traditional automatic segmentation methods, such as Bruch's membrane and the choroid-scleral junction, as well as to methods for automatically identifying and correcting errors in retinal layer segmentation. [Background technology]

[0002] OCT is a noninvasive imaging technique that uses light waves to generate cross-sectional images of retinal tissue. For example, OCT allows for the visualization of distinct tissue layers of the retina. Generally, OCT systems are interferometric imaging systems that determine the scattering profile of a sample along the OCT beam by detecting the interference of light reflected from the sample with a reference beam, forming a three-dimensional (3D) representation of the sample. Each scattering profile in the depth direction (e.g., z-axis or axial direction) can be individually reconstructed into an axial scan or A-scan. Cross-sectional two-dimensional (2D) images (B-scans) and extended 3D volumes (C-scans or cube scans) can be constructed from multiple A-scans acquired as the OCT beam is scanned / translated through a set of transverse (e.g., x-axis and y-axis) locations on the sample. OCT also allows for the construction of planar, en face (e.g., en face) 2D images of selected portions of a tissue volume (e.g., a target tissue slab (subvolume) or target tissue layer(s) of the retina). Optical Coherence Tomography Angiography (OCTA) is an extension of OCT that can identify (e.g., render in an image format) the presence or absence of blood flow in tissue layers. OCTA can identify blood flow by identifying differences (e.g., contrast differences) over time in multiple OCT images of the same retinal region and designating differences that meet predetermined criteria as blood flow. A more detailed description of OCT and OCTA is provided below.

[0003] Identifying retinal layers in OCT data is often beneficial for diagnostic purposes to better observe specific tissues. Identifying retinal layers in OCT data allows for focusing on specific portions of a B-scan or for better forming en face images based on selected retinal layers. Manual segmentation of retinal layers can be very time-consuming and inconsistent. Therefore, while automated retinal layer segmentation tools are important for segmenting retinal layers in OCT data, the reliability of such tools is compromised by a decline in OCT data quality and / or the emergence of pathologies that can alter the typical (e.g., expected) shape of retinal layers. Therefore, the performance of multilayer segmentation methods / tools is an important determining factor when evaluating such tools, especially in the presence of structure-altering retinal pathologies (such as those caused by various retinal diseases).

[0004] Ideally, automated multilayer segmentation methods would determine the desired layer boundaries without manual intervention. However, these methods are prone to layer misidentification errors, especially in eyes with moderate to severe retinal pathology and in low-quality OCT data. Layer boundaries in such cases are usually not identifiable automatically or even manually. While layer misidentification can be confounded by OCT imaging errors such as poor signal quality, eye movement, and decentration, it is largely caused by the morphological complexity and reflectivity changes of retinal structures in disease cases. Accurate multilayer segmentation in such cases exceeds the capabilities of conventional segmentation algorithms.

[0005] Manual multi-layer segmentation by a human rater typically requires the rater to identify layers by freehand drawing or by placing points that are used in an interpolation or fitting algorithm. Manual methods are more time-consuming, labor-intensive, and have higher inter-rater variability than automated methods. Manual segmentation may be impossible due to missing retinal layers or boundaries.

[0006] The difficulty of automated retinal layer segmentation is particularly evident when attempting to segment retinal layers that are typically not clearly defined in OCT data (e.g., structural data), such as the Bruch's membrane (BM) or choriocapillaris layer. Because the choriocapillaris layer is very thin, creating an en face image for visualizing 3D OCTA data is generally effective for elucidating choriocapillaris features. Segmentation of the retinal pigment epithelium (RPE) in cases of age-related macular degeneration (AMD) accompanied by drusen (subretinal lipid deposits) or pigment epithelial detachment (PED) generally cannot be used to generate an en face image of the choriocapillaris. Therefore, accurate BM segmentation can be key to elucidating choriocapillaris features. Unfortunately, segmenting the BM can be difficult due to reasons such as the fact that structural OCT data typically have low contrast in the BM region, the decorrelation tail problem between the RPE and BM in OCTA data, and / or signal distortion around the BM in pathological cases. Combining OCT structural data with OCT flow data to identify the BM has been proposed in Non-Patent Document 1. While the details of the method are not immediately clear, Non-Patent Document 1 appears to use an automated OCT-OCTA graph cut algorithm for segmenting Bruch's membrane in the presence of drusen. [Prior art documents] [Non-patent literature]

[0007] [Non-Patent Document 1] Schottenhamml et al., "OCT-OCTA segmentation: a novel framework and a application to segment Bruch's membrane in the presence of drusen," Invest Ophthalmology and Visual Science, 2017, 58(8), 645-645 Summary of the Invention

[0008] It is an object of the present invention to provide an automated method / system that provides more reliable segmentation of previously difficult to segment retinal layers such as Bruch's membrane and the choriocapillaris.

[0009] Another object of the present invention is to provide an automated method / system for retinal layer segmentation in the presence of pathology. It is a further object of the present invention to provide an automated method / system for identifying errors in retinal layer segmentation.

[0010] It is yet another object of the present invention to provide an automated method / system for replacing errors in retinal layer segmentation with approximations. The above objectives are achieved in a method / system for automated segmentation of Bruch's membrane (BM) and other retinal layers in optical coherence tomography (OCT) data.

[0011] OCT structural data is fundamentally different from OCTA flow data because they provide different types of information. Therefore, their individual images can be quite different, especially in the upper retinal layers where OCT typically provides good structural information. In these upper layers, the OCT and OCTA data can be quite different. OCT data generally lacks definition in the lower retinal layers. However, applicants have discovered that in these lower retinal layers, the OCT data can appear similar to the OCTA data. The present invention utilizes this discovery to better highlight / distinguish the transition from when the OCT data differs significantly from the OCTA data to when the OCT data becomes more similar to the OCTA data. In particular, the choroid and sclera regions are similar in the structural OCT and flow OCTA images. By noting where the OCT data is similar to the corresponding OCTA data in the slab, the choroid and sclera regions can be removed or attenuated from the structural OCT (or otherwise defined / demarcated in the structural OCT) (using the corresponding OCTA data). In this approach, structural OCT data around Bruch's membrane can be enhanced after attenuation of the choroidal region. Therefore, the enhanced OCT data greatly simplifies the segmentation problem. This approach is particularly beneficial in the lower retinal layers, but can also be applied to other target retinal layers.

[0012] The present invention improves retinal layer segmentation in OCT data by using OCTA data to enhance regions at or around a target retinal layer (e.g., BM, choroid-sclera junction, etc.) in the OCT data that may otherwise lack sufficient definition for segmentation. The OCT data may be enhanced based on blending the OCT data with OCTA data. In one example, the contrast of the OCT data may be enhanced in regions where the OCT and OCTA data are dissimilar (e.g., above the target retinal layer) and reduced in regions where the OCT and OCTA data are similar (e.g., below the target retinal layer). In another example, the contrast in the OCT data may be enhanced around a retinal layer of interest (e.g., Bruch's membrane) by utilizing the similarities and dissimilarities between OCT and OCTA. Enhancing the OCT data may include subtracting a percentage of the (e.g., weighted) mixture of the OCT data and the OCTA data from the OCT data, where the percentage may be based on a ratio of a measure of joint variability (e.g., statistical covariance) of the OCT data and the OCTA data to a measure of data spread (e.g., statistical variance) of the OCT data and the OCTA data. Thus, the enhanced OCT data may enhance the boundary of the target retinal layer. The target retinal layer in the OCT data may then be segmented based on the enhanced data. Other layers may then be segmented relative to the target retinal layer.

[0013] While the present invention provides improved automatic retinal segmentation, any segmentation method is prone to errors. Therefore, the present invention also provides error detection and correction or approximation of automatic segmentation. That is, the present invention provides a method for identifying unsuccessful segmentations of retinal layers and replacing them with segmentation approximations. In certain embodiments, unsuccessful segmentations in OCT structural data are automatically identified using OCTA angiography (OCTA) retinal layer slabs (e.g., OCTA en face images). Two en face images of the OCTA data containing the target retinal layer (to be checked for unsuccessful segmentations) are used to check for errors in the segmentation of the target retinal layer in the OCT data. For example, the target retinal layer can be the inner plexiform layer (IPL), the outer plexiform layer (OPL), etc. Alternatively, the first en face image and the second en face image can be formed from separate slabs from the OCT data.

[0014] The determination of the failure or success of the segmentation of the target retinal layer is based on a local or global similarity measure between the first en face image and the second en face image. For example, the similarity may be based on the normalized cross-correlation (NCC) between the two en face images. The identified local or global error may be automatically replaced with an approximation based on the location of the topmost and bottommost retinal layers of one of the two en face images. For example, the target retinal layer is sandwiched between the topmost and bottommost layers of the slab forming the second en face image (the slab forming the first en face image may have the same top layer but the bottom layer defined by the target retinal layer), and the approximation may be based on a weighted combination of the topmost and bottommost layers of the slab forming the second en face image. The topmost and bottommost layers of the second en face image may be selected based on a confidence measure that the topmost and bottommost layers are error-free. For example, the top and bottom layers may be selected based on a light-to-dark or dark-to-light sharpness transition measure in the OCT data, and the weighted combination may be based on the location of the top and bottom layers relative to the expected location of the target retinal layer sandwiched between them.

[0015] Other objects and achievements of the present invention, together with a fuller understanding of the invention, will become apparent and understood by reference to the following description and claims taken in conjunction with the accompanying drawings.

[0016] To facilitate the understanding of the present invention, several publications are cited or referenced herein. All publications cited or referenced herein are incorporated by reference in their entirety.

[0017] The embodiments disclosed herein are merely examples, and the scope of the present disclosure is not limited thereto. Features of any embodiment described in one claim category, e.g., a system, may also be claimed in other claim categories, e.g., a method. Dependencies or back-references in the appended claims are selected for formality reasons only. However, any subject matter available from a careful back-reference to a previous claim may also be claimed, thereby disclosing any combination of claims and their features and may be claimed regardless of the dependencies selected in the appended claims. [Brief explanation of the drawings]

[0018] The patent or application file contains at least one drawing executed in color. Copies of any color drawing(s) of this patent or patent application publication will be provided by the Office upon request and payment of the necessary fee.

[0019] In the drawings, like reference symbols / letters refer to like elements. [Figure 1] FIG. 1 provides examples of the seven retinal layers that may typically be segmented by state-of-the-art automated multi-layer segmentation tools. [Figure 2] FIG. 1 shows a structural OCT B-scan of an eye with geographic atrophy (GA) and the corresponding OCTA B-scan. [Figure 3] FIG. 1 shows a structural OCT B-scan of an eye with age-related macular degeneration (AMD) and the corresponding OCTA B-scan. [Figure 4] FIG. 1 illustrates the application of a two-stage segmentation process to enhance data in accordance with the present invention. [Figure 5A] FIG. 1 shows two examples of enhanced OCT data for BM visualization using structural and flow data. [Figure 5B] FIG. 1 shows three examples of BM segmentation with corresponding choriocapillaris vasculature maps. [Figure 6]FIG. 1 shows the mean absolute difference (with 95% confidence intervals) and R2 between two readers and between a reader and the BM segmentation. [Figure 7] Choroidal thickness maps (in microns) of a right eye with an overlaid ETDRS grid centered on the fovea, and a structural choroidal vasculature map (right), created using manual segmentation (left) and automated segmentation (center). [Figure 8] FIG. 1 provides a table showing information extracted from regression and Bland-Altman analyses for each sector of the ETDRS grid. [Figure 9] 10A-10C illustrate an exemplary ILM-IPL en face angiography slab image and an exemplary ILM-OPL en face angiography slab image defined using the segmentation algorithm output from the automated segmentation tool. [Figure 10] FIG. 10 provides a plot showing the normalized cross-correlation (NCC) of all rows (B-scans) compared in the en face image of FIG. 9. [Figure 11] FIG. 10 shows a B-scan and an ILM-IPL en face slab image in which the IPL segmentation of the volume data has been replaced by an IPL segmentation approximation. [Figure 12] FIG. 7 shows two examples of an SRL slab using MLS IPL segmentation 73a and an IPL approximation 73b superimposed on two B-scans with incorrect MLS IPL segmentation. [Figure 13] FIG. 1 shows an image (FOV=8 mm) with unsuccessful automatic segmentation at the IS / OS junction. [Figure 14] FIG. 1 shows an image (FOV=16 mm) with unsuccessful automatic segmentation in the foveal region. [Figure 15]FIG. 20 shows an example of an OCT B-scan (FOV=12 mm) with overlapping segmentations (top to bottom) depicting the boundaries of various retinal layers, similar to that shown in FIG. 19. [Figure 16] FIG. 1 shows a schematic process of an automated method based on segmentation propagation according to the present invention. [Figure 17] FIG. 1 illustrates a semi-automated method based on segmentation reflection. [Figure 18] FIG. 1 illustrates a generalized frequency-domain optical coherence tomography system used to collect 3D image data of the eye suitable for use in the present invention. [Figure 19] FIG. 1 shows an exemplary OCT B-scan image of a normal retina of a human eye, illustratively identifying various normal retinal layers and boundaries. [Figure 20] FIG. 10 shows an example of an en face vascular image. [Figure 21] FIG. 1 shows an exemplary B-scan of a vasculature (OCTA) image. [Figure 22] FIG. 1 illustrates an exemplary computer system (or computing device or computer). DETAILED DESCRIPTION OF THE INVENTION

[0020] Accurate detection of anatomical and pathological structures in optical coherence tomography (OCT) images is important for the diagnosis and research of retinal diseases. Manual segmentation of features of interest in each B-scan of an OCT volume scan not only requires skilled evaluators but is also very time-consuming for clinical use. Another problem is the inherent inter-evaluator variability, which leads to subjective segmentation results. A fully automated method for segmenting multiple retinal layer boundaries in B-scans could significantly reduce the processing time required for segmentation.

[0021] Using automated multilayer segmentation techniques to segment retinal layers has several advantages. For example, such tools can eliminate redundant preprocessing steps such as noise reduction, resampling and normalization, and OCT cube flattening. Furthermore, additional information can be determined based on the identified segmentation layers, such as referencing an unknown layer to one or more known layers, identifying a layer by noting two adjacent layers that bound it, or identifying smaller regions for processing. Automated multilayer segmentation tools can also enable the implementation of other analysis tools, such as various thickness maps (macula, RNFL, and ganglion cell thickness) and en face imaging, such as structural and angiographic en face images. This can also serve as input to other algorithms, such as fovea finder, OCTA decorrelation tail removal, and CNV finder algorithms.

[0022] Figure 1 provides examples of seven retinal layers that may typically be segmented by state-of-the-art automated multi-layer segmentation tools. While current tools can successfully automatically segment the upper retinal layers, there are additional retinal layers, such as Bruch's membrane and the choroid region, that typical automated multi-layer segmentation tools cannot reliably segment or identify, as described below with reference to Figure 19. These layers are typically lower layers where the OCT signal is weaker and may be prone to higher levels of artifacts, and / or layers with pathologies that alter the typical retinal layer structure of a healthy retina.

[0023] Therefore, the performance of multilayer segmentation tools becomes a critical determinant in the presence of structure-altering retinal pathologies caused by various retinal diseases. Existing automated multilayer segmentation tools have two main problems. First, they are prone to layer segmentation errors, especially in eyes with moderate to severe retinal pathologies. Second, most existing methods are computationally very expensive, requiring minutes to hours to compute retinal layer segmentations on large OCT data cubes. Layer segmentation errors can be confounded by OCT imaging errors, such as poor signal quality, eye movement, and disease-induced morphological complexity and reflectance changes in retinal structures. Multilayer segmentation in these cases exceeds the capabilities of conventional segmentation algorithms.

[0024] The present invention provides a method and system for automatically segmenting retinal layers that are not typically included in automated segmentation tools, such as Bruch's membrane (BM) and choroid. To better segment these layers, the present invention uses corresponding OCTA B-scans to enhance the contrast of structural OCT B-scans around the BM (or other target retinal layers) by removing or attenuating the OCT signal below the BM, such as the choroid and scleral portions, without using any prior segmentation.

[0025] Figure 2 shows a structural OCT B-scan 11 of an eye with geographic atrophy (GA) and the corresponding OCTA B-scan 13. The OCT data 11 and OCTA data 13 are blended, as indicated by block 15, and the resulting blended signal is then subtracted from the OCT data 11, as indicated by block 17, to generate enhanced structural data / image 19. The enhanced structural image 19 shows much higher contrast, especially in the GA region, which enables BM segmentation even within this pathology region. BM segmentation using OCT or OCTA without normalization is a challenging problem due to the low contrast in OCT around the BM and the decorrelation tail in OCTA, and no clear junction would be available at the periphery of the BM layer.

[0026] 3 shows a structural OCT B-scan 21 of an eye with age-related macular degeneration (AMD) and a corresponding OCTA B-scan 23. Again, the OCT data 21 and OCTA data 23 are blended as indicated by block 25 and subtracted (e.g., block 27) from the OCT data 21 to generate an enhanced structural image / data 29. As shown, the enhanced structural image 29 exhibits higher contrast below the retinal pigment epithelium (RPE) retinal layer.

[0027] The generation of the enhanced structural image 19 / 29 may constitute all or part of the first stage of a two-stage (or more) image segmentation process. That is, after the enhanced OCT data is generated in the first stage, any suitable segmentation method / algorithm may then be applied to the enhanced OCT data (e.g., 19 and / or 29) in one or more subsequent stages to provide automated (or semi-automated) segmentation.

[0028] This specification presents several general frameworks and methods for automatic and semi-automatic multi-layer segmentation. One such automatic segmentation method, described below in connection with Figures 13-17, is relatively fast and can be used in commercial products. This approach automatically or manually identifies a starting location (e.g., a reflected starting location and / or B-scan) and then reflects multi-layer segmentation information, for example, to adjacent B-scans and / or nearby B-scans (e.g., B-scans within a predetermined distance from the current location (e.g., another B-scan from the starting location)). Another segmentation approach suitable for use with this two-stage (or more-stage) image segmentation method can use a graph search algorithm based on contrast-enhanced structural OCT data. Regardless of the segmentation method used, each stage can apply segmentation to different resolutions of the enhanced (OCT) data. That is, the enhanced (OCT) data may be downsampled to different resolution levels, and each resolution level may be provided to a different stage of the segmentation process, with the first stage of the segmentation process segmenting the lowest resolution, and the output of each stage being the starting segmentation for the next stage, where the resolution of the next stage is equal to or higher than the previous stage.

[0029] In this exemplary two-stage (or more) segmentation process, in the first stage, the selected image segmentation method being used produces a preliminary, coarse segmentation result. Then, by taking the baseline of the first-stage segmentation as the initial segmentation, the second stage of the segmentation process can begin by segmenting below the baseline (using any appropriate segmentation method) to obtain the final segmentation result. With proper initialization (e.g., initial segmentation) in the first stage, the second stage (and subsequent stages) can achieve the desired segmentation result, even for difficult images.

[0030] 4 illustrates the application of this two-stage segmentation process to the augmented data. An embodiment of the present invention may include the following steps. 1) Enhance the structural OCT by removing or attenuating the choroidal region in the structural OCT volume using the corresponding OCTA volume.

[0031] V e =V s -α(w1V s +w2V a ) α^=argmin a f((w1V s +w2V a ),V s -α(w1V s +w2V a )) V s :Structural OCT volume V a :OCTA volume V e :Enhanced structural OCT volume w1,w2: weights for OCT and OCTA volumes f: Objective function (e.g., squared normalized cross-correlation, mutual information) α: Parameter for optimization The solution is α^=Cov(w1V s +w2V a ,V s ) / Var(w1V s +w2V a ) where the objective function f is the square of the normalized cross-correlation, Cov is the covariance, and Var is the variance.

[0032] 2) The first step as initial segmentation is V with high confidence. e It consists of segmenting each B-scan of the image, followed by calculating a baseline, which is used for the second stage of segmentation.

[0033] In FIG. 4, image 31 is an augmented image with high confidence (e.g., V e ) and the baseline 31b calculated from the 2D fit.

[0034] 3) The second stage consists of the final segmentation by segmenting below the initial segmentation. In FIG. 4, the image 33 is an augmented version (e.g., V e ) shows the final segmentation 35. This segmentation can then be transplanted back into the original OCT data (e.g., structural B-scan), as shown in image 37.

[0035] In this embodiment, the first and second stage segmentation methods are graph search algorithms, although other segmentation methods may be used. As described above, the present invention can be used in a method for automated Bruch's membrane segmentation in optical coherence tomography (OCT). Accurate Bruch's membrane (BM) segmentation is essential for characterizing possible choriocapillaris defects and retinal pigment epithelial prominence and dysfunction, which are important diagnostic indicators of retinal disease. This BM segmentation method / system can be applied to OCT volumes.

[0036] This exemplary BM segmentation method uses a structural OCT volume (V s ) and flow OCT volume (V a ) and enhance the BM layer by using the enhanced OCT volume (V e ) is V e =V s -α(w s V s +w a V a ), where w s and w a is the weight (for example, set to 0.5). The proportional coefficient α ise and mixed (w s V s +w a V a ) is minimized, and the similarity (squared normalized cross-correlation) between s V s +w a V s ,V s ) / Var(w s V s +w a V a This segmentation method is based on a multiresolution approach and a graph search algorithm. The segmentation baseline at each resolution level is used as the starting segmentation for the next higher-resolution segmentation. In this example, the number of resolution levels is set to two for faster processing. The performance of the algorithm is evaluated by comparing it with manual editing from two readers using 120 B-scans extracted from prototype OCTA cube scans of 40 eyes, measuring 3 x 3 mm, 6 x 6 mm, 9 x 9 mm, 12 x 12 mm, and 15 x 9 mm, acquired using a 200 kHz PLEX® Elite 9000 (ZEISS, Dublin, CA). All scans were a mixture of disease cases, including DR and AMD.

[0037] Figure 5A shows the structural data (V s ) and flow data (V a ) for BM visualization using enhanced OCT data (V e ) and FIG. 5B shows three examples 43a, 43b, and 43c of BM segmentations 45a, 45b, and 45c with corresponding choriocapillaris vasculature maps 47a, 47b, and 47c. Examples 41a and 41b show the corresponding structural OCT (V s ) B-scan and flow OCTA (V a ) B-scan of enhanced OCT (V eExamples 43a, 43b, and 43c show segmentation results 45a, 45b, and 45c with corresponding choriocapillaris slabs 47a, 47b, and 47c by using OCTA flow data (e.g., generated OCTA slabs) between the BM and BM+20 μm.

[0038] Figure 6 shows the mean absolute difference (with 95% confidence intervals) and R between the two readers and between the reader and subject BM segmentations. 2 The mean absolute difference for each scan pattern demonstrates strong correlation and great agreement between readers and BM segmentation.

[0039] Overall, the automated and manual segmentations have strong correlation and great agreement. Automated segmentation can be a valuable diagnostic tool for retinal diseases.

[0040] Another exemplary embodiment demonstrates a method for segmenting the choroid-scleral junction in optical coherence tomography (OCT). A relatively fast algorithm was developed to segment the choroid-scleral junction. The segmentation begins with B-scans with high contrast around the choroid-scleral boundary. The segmentation is then propagated throughout the volume data. The algorithm uses intensity and gradient images as input to a graph-based method to segment each B-scan at a region of interest. The performance of the algorithm was evaluated using normal SS-OCT volume data from 49 eyes, consisting of 500 x 500 A-scans covering 12 x 12 mm, acquired using a PLEX® Elite 9000 SS-OCT (ZEISS, Dublin, CA). Choroidal thickness maps, defined as the distance between the fitted RPE baseline and the choroid-scleral junction, were generated using both manual and automated segmentation. The performance of this embodiment is reported using regression and Bland-Altman analyses for each sector of the ETDRS grid. Figure 7 shows choroidal thickness maps (in microns) for a right eye with an overlaid ETDRS grid centered on the fovea, created using manual segmentation (left) and automatic segmentation (center), along with a structural choroidal vasculature map (right). The ETDRS grid consists of three concentric circles with radii of 2 mm, 4 mm, and 6 mm centered on the fovea. The choroidal thickness map in Figure 7 is created by manual and automatic segmentation, as well as the corresponding structural choroidal vasculature map.

[0041] Figure 8 provides a table showing the information extracted from the regression and Bland-Altman analyses for each sector of the ETDRS grid. The regression and Bland-Altman analyses for all sectors demonstrate strong correlation and good agreement between the manual and automated methods. The average processing time is less than 4 seconds using an Intel i7 CPU, 2.7 GHz, and 32 GB memory. Overall, the choroidal thickness maps produced by the automated and manual segmentations have strong correlation and good agreement.

[0042] As shown, the present invention provides good results in an automated segmentation system. However, as mentioned above, automated segmentation systems can sometimes produce erroneous results due to numerous factors that are generally beyond the control of the automated segmentation system. For example, segmentation errors can occur due to poor OCT signal quality, eye movement, or morphological complexity and reflectance changes of retinal structures in disease cases. In light of such problems associated with automated (and manual) segmentation systems, the present invention also proposes a method for identifying unsuccessful segmentations of retinal layers and replacing them with segmentation approximations.

[0043] Traditionally, segmentation quality is determined using the segmentation confidence at each layer point. The segmentation confidence is usually determined based on the intensity of the cost image (e.g., gradient image) at a given segmentation point. This method can sometimes not work well because one segmentation may jump to an adjacent layer segmentation and still have a high confidence.

[0044] In this invention, OCTA flow data can be used to determine the segmentation quality of OCT structural data. Alternatively, OCT data can also be used for this proposal. In this example, the similarity of OCTA vasculature slabs can be used as an indicator of the failure of a particular layer segmentation. Segmentation failures can be locally identified and replaced by approximations.

[0045] This embodiment automatically identifies unsuccessful segmentation using angiographic retinal layer slabs. For example, unsuccessful layer segmentation of the inner plexiform layer (IPL) can be detected using the inner limiting membrane (ILM) and outer plexiform layer (OPL) layers, and this detection is performed to generate ILM-IPL and ILM-OPL angiographic (or structural) slabs, assuming that the IPL and ILM segmentations are reasonably accurate. Using ILM and OPL segmentation for this purpose typically works better than using the retinal nerve fiber layer (RNFL), IPL, and inner nuclear layer (INL). This is due to the clearer transitions from various parts of the ILM (dark areas) to the RNFL (light areas) and from the OPL (light areas) to the avascular zone (dark areas). In this example, if the IPL segmentation is performed appropriately, a high degree of local similarity between the ILM-IPL angiography slab and the ILM-OPL angiography slab is expected. Local similarity is an indicator of unsuccessful IPL segmentation. The slabs can be generated based on maximum projection (or other suitable methods for forming en face images) within defined layer boundaries in the OCTA volume data. If the IPL segmentation is unsuccessful, the IPL segmentation is replaced with an IPL approximation based on a weighted average of the ILM and OPL. The same approach can be used to identify unsuccessful OPL segmentation (or unsuccessful OPL segmentation of other target retinal layers) by appropriate selection of the reference layer segmentation. For example, the IPL segmentation or the IS / OS segmentation can be used as the reference layer segmentation if the segmentation of these layers is correct.

[0046] In summary, this embodiment can automatically identify the failure of local segmentation of retinal layer boundaries using angiographic retinal layer slabs.In the following example, the failure of IPL layer segmentation is detected by using ILM layer and outer OPL layer to generate ILM-IPL and ILM-OPL angiographic slabs, under the assumption that IPL and ILM segmentation are acceptable.For this purpose, ILM and OPL segmentation usually performs better than other inner retinal layer segmentation (such as RNFL, IPL, and INL), because the transitions from various parts (dark areas) to RNFL (light areas) and from OPL (light areas) to avascular zone (dark areas) are sharper for ILM.

[0047] If IPL segmentation is performed appropriately well, a high degree of local similarity is expected between the ILM-IPL angiography en face image (or slab) and the ILM-OPL angiography en face image (or slab). Local similarity is an indicator of unsuccessful IPL segmentation. In this example, the en face slab image is generated based on maximum projection within the defined layer boundary in the OCTA volume data. If the IPL segmentation is unsuccessful, the IPL segmentation can be replaced with an IPL approximation based on a weighted average of the ILM and OPL.

[0048] FIG. 9 shows an exemplary ILM-IPL en face angiography slab image 51 and an exemplary ILM-OPL en face angiography slab image 53 formed using the segmentation algorithm output from the automated segmentation tool. As shown, the majority of en face images 51 and 53 are similar, indicating no segmentation failures. However, a local region 55 in en face image 51 is dissimilar to the corresponding local region 55 in en face image 53, indicating an unsuccessful segmentation of the IPL in this local region. The reason for the dissimilarity is believed to be that the slab in image 51 was formed based on an automated IPL segmentation, and the unsuccessful segmentation may be due to the slab in image 51 not conforming to the true shape of the IPL (e.g., the lowest layer forming this slab). The similarity between the rows of the angiographic slabs of ILM-IPL51 and ILM-OPL53 (e.g., corresponding to each B-scan) can be measured by the normalized cross correlation (NCC) between the rows of these two slabs.

[0049] FIG. 10 provides a plot showing the NCC of all rows (B-scans) numbered from 0 at the top to 1200 at the bottom. As can be seen from the plot, the similarity in the upper part (e.g., a contiguous local region) of the angiographic slab (e.g., B-scans 50 to 200) is low relative to other parts of the angiographic slab. This is an indication of unsuccessful IPL segmentation in this upper part. If the NCC of a B-scan is less than a predetermined threshold (e.g., 0.5 to 0.7, or other suitable value), it is determined that unsuccessful IPL segmentation occurred in that B-scan.

[0050] The IPL segmentation of this unsuccessful region can be replaced by an IPL segmentation approximation based on a weighted average of the ILM and OPL (e.g., 0.4*ILM+0.6*OPL). For example, the weights of the ILM (top layer) and OPL (bottom layer) can be based on their relative positions to the expected position of the target retinal layer. Other layers can be determined / approximated / checked based on the IPL approximation. The RNFL segmentation approximation can be calculated based on a weighted average of the ILM and IPL approximations (e.g., 0.8*ILM+0.2*IPL_approx). The INL segmentation approximation can be calculated based on a weighted average of the OPL and IPL approximations (e.g., 0.6*OPL+0.4*IPL_approx).

[0051] Alternatively, if the variance of the NCC functions for all B-scans is less than a threshold, the IPL segmentation of the volume data can be replaced with an IPL segmentation approximation. Figure 11 shows a B-scan 61 and an ILM-IPL en face slab 63, in which the IPL segmentation of the volume data has been replaced with an IPL segmentation approximation. In B-scan 61, the solid and dashed lines represent the segmentation and approximation results for the RNFL, IPL, and INL, respectively. Unsuccessful local segmentation can also be determined by finding the similarity between the ILM-IPL angiography slab and the ILM-OPL angiography slab in a window spanning a portion of the B-scan (or a portion of the B-scans), rather than the entire B-scan as described above.

[0052] In summary, in this embodiment, OCT angiography is used to identify segmentation failures. The process may use two or more reference layer segmentations to identify local segmentation failures.

[0053] In an example implementation, the present invention was used to provide automated inner retinal layer segmentation approximation for cases of advanced retinal disease in optical coherence tomography angiography (OCTA).

[0054] Generally, automated multilayer segmentation (MLS) methods determine desired inner layer boundaries. However, they are prone to layer misidentification errors, especially in eyes with retinal pathology and low-quality data. In such cases, inner layer boundaries often cannot be manually identified. This example provides an automated inner plexiform layer (IPL) outer boundary segmentation approximation method using OCTA volumes.

[0055] MLS detects unsuccessful IPL segmentation using internal limiting membrane (ILM)-IPL angiographic slabs and ILM-outer plexiform layer (OPL) angiographic slabs generated based on the outer boundaries of the ILM and OPL segmentations. This assumes that the ILM and OPL segmentations are correct. The local similarity between the ILM-IPL and ILM-OPL angiographic slabs, as measured by normalized cross-correlation (NCC), is expected to be low if the MLS IPL segmentation fails when these slabs are generated based on maximum projection. If the variance of the NCC is less than a threshold, the MLS IPL segmentation is replaced by an IPL segmentation approximation as a weighted average of the ILM and OPL segmentations; otherwise, the MLS IPL segmentation is used.

[0056] The performance of this example was evaluated using 161 angiographic volumes spanning 3x3mm (76 scans), 6x6mm (67 scans), 8x8mm (2 scans), 12x12mm (7 scans), HD6x6mm (6 scans), and HD8x8mm (3 scans) acquired using the CIRRUS™ HD-OCT 161 with AngioPlex® OCT Angiography (ZEISS, Dublin, CA). The data represented a mix of retinal disease. Clinical raters rated each superficial retinal layer (SRL) slab generated by the new algorithm as successful or unsuccessful.

[0057] Figure 12 shows two examples of pre- and post-correction SRL slabs 71 and 72 (e.g., original slabs 71A / 72A with erroneous MLS segmentation and corrected slabs 71B / 72B). Both slabs 71 and 72 show the MLS IPL segmentation 73a (e.g., a dashed cyan line in the color image or a dotted line in the monochrome image) and the IPL approximation 73b (e.g., a solid cyan line in the color image or a dashed line in the monochromatic image) superimposed on two B-scans with incorrect MLS IPL segmentation. As shown in the corrected slabs 71B and 72B, the IPL approximation forms a more accurate representation of the SRL in these cases. The MLS IPL segmentation was replaced with the IPL approximation in 34 (21%) of the scans. The success rate of MLS was 79% without the IPL approximation and 96% with the approximation. Thus, the present example produced acceptable SRL slabs when MLS IPL segmentation performance encountered issues with severe retinal disease or poor image quality.

[0058] Presented herein is an alternative exemplary segmentation method described above that automatically or manually identifies a starting location (e.g., a reflection starting location and / or B-scan) and reflects multi-layer segmentation information. For illustrative purposes, FIG. 13 shows an image (FOV=8 mm) with unsuccessful automatic segmentation at the IS / OS junction, which also affected the RPE segmentation. By manually correcting the IS / OS segmentation and then reflecting, the RPE can be accurately and automatically resegmented. Similarly, FIG. 14 shows an image (FOV=16 mm) with unsuccessful automatic segmentation in the foveal region, which affected the inner retinal layers as well as the IS / OS and RPE segmentation. Again, manual correction in the foveal region followed by reflection can correct the segmentation in this region.

[0059] This embodiment is an automated method based on propagation of multi-layer segmentation throughout the OCT volume, starting with the B-scan with the best intraretinal / extraretinal contrast. This contrasts with conventional propagation methods, which are semi-automated and limited to propagation along a single layer boundary. This embodiment can use a semi-automated method similar to this automated method, except that the starting B-scan can be selected and / or partially edited by an expert. In the semi-automated method, it is possible to edit some layer boundaries of the starting B-scan. The automated portion of the algorithm can then further segment the remaining unedited layer boundaries before propagation.

[0060] The automated method is relatively fast and therefore suitable for use in commercial applications. It is based on the concept of multi-layer segmentation reflection (e.g., reflection is based on simultaneous multi-layer boundaries). The automated method naturally starts from a portion of the retina scan with healthy structure, as in the normal case, or from enhanced OCT data (image), or from other suitable high-quality portions of the image (OCT data). Starting the segmentation (B-scan) from a healthy portion of the retina and smoothly transitioning to adjacent B-scans makes the reflection algorithm relatively fast and robust.

[0061] This document describes the general concept of automatic segmentation reflection. A pre-processing method suitable for use in the present invention is described in commonly assigned U.S. Patent No. 10,169,864. However, the actual segmentation and reflection workflow differs.

[0062] FIG. 15 shows an example of an OCT B-scan (FOV=12 mm) with overlapping segmentations (top to bottom) depicting the boundaries of various retinal layers (similar to those shown in FIG. 19 ), including the internal limiting membrane (ILM), the outer boundary of the retinal nerve fiber layer (RNFL or NFL), the outer boundary of the inner plexiform layer (IPL), the outer boundary of the inner nuclear layer (INL), the outer boundary of the outer plexiform layer (OPL), and the IS / OS junction, and the retinal pigment epithelium (RPE).

[0063] Layer boundaries with a positive axial gradient (dark-to-light transition) generally include the boundary between the vitreous and ILM, the upper boundary of the IS / OS-associated bright line, and the lower boundary of the INL-associated bright line. Layers with a negative axial gradient (light-to-dark transition) generally include the outer boundary of the RNFL, IPL, and OPL (and the outer boundary of the BM).

[0064] For example, layer boundaries with positive axial gradients can be segmented simultaneously or one by one in a B-scan. These segmentations can serve as a baseline for segmentation of adjacent B-scans. These segmentations can also define regions of interest in adjacent B-scans for segmentation of layer boundaries with either positive or negative axial gradients. Layer boundaries with negative axial gradients can be segmented simultaneously or one by one in a B-scan.

[0065] FIG. 16 shows a schematic process of the automated method based on segmentation reflection according to the present invention. First, in step 91, an initial segmentation is performed. Within this step, a subset of B-scans with the highest contrast around a layer or layers of interest are automatically selected from the OCT volume. A segmentability metric, such as that described in U.S. Pat. No. 9,778,021 (assigned to the same assignee as the present invention), may be used to determine the local contrast of the B-scans. The subset of B-scans may be segmented independently or dependently (if they are spatially close to each other) using a graph-based method or other fast methods known in the art. In step 93, a starting B-scan is selected, along with a starting segmentation to be reflected. A segmentation confidence measure may be used to determine which B-scan segmentation from the subset of B-scans is most reliable. Preferably, each segmentation point is assigned a confidence value, which may be a value obtained from a total cost function (image) at the segmentation point. The total cost image may be used in graph-based segmentation to locate the layer of interest. In the third step 95, segmentation is propagated starting from the B-scan selected in the previous step. The segmentation is propagated to B-scans adjacent to the starting point. In this process, the segmentation line of a B-scan becomes the baseline for the same-layer segmentation of adjacent B-scans. It is assumed that the change in segmentation between two adjacent B-scans is defined within (e.g., confined within) a (predefined) range used to extract the region of interest in the B-scan to be segmented (e.g., the segmentation is assumed to be limited within this region of interest). For example, the range may be approximately 1 to 5 times the spacing between two adjacent B-scans. For example, if the spacing between two adjacent B-scans is 12 microns, the range may be 10 to 30 microns.Alternatively, each segmentation line is bounded by two adjacent segmentation lines, so that the region (e.g., region of interest) can be the area between the two adjacent segmentation lines. For example, a first B-scan_1 is segmented, followed by a second B-scan_2, which uses the segmentation of B-scan_1 as a reference to define the OCT region of interest for layer segmentation. Once B-scan_2 is segmented, B-scan_2 becomes the reference for a third B-scan_3.

[0066] FIG. 17 illustrates a semi-automated method based on segmentation reflection. Semi-automated segmentation and reflection for a layer boundary or layers replaces steps 91 and 93 in the automated method of FIG. 16 with expert input. In block 101, the segmentation line (e.g., determined by the automated method) is corrected in selected B-scans where unsuccessful segmentation is observed. In block 103, the segmentation line is reflected in a manner similar to step 95 in the automated method of FIG. 16. As shown in block 105, the segmentation line is either reflected throughout the entire OCT volume (if a segmentation determined by the automated method is not available) or stopped for a layer boundary if the error between the automated segmentation and the current segmentation is less than a threshold value (e.g., a 5-micron threshold). Blocks 101 through 105 are then repeated until all segmentation lines in the OCT volume have been corrected, as shown by block 107.

[0067] The following describes various hardware and architectures suitable for the present invention. Optical coherence tomography system Generally, optical coherence tomography (OCT) uses low-coherence light to generate two-dimensional (2D) and three-dimensional (3D) internal views of biological tissues. OCT enables in vivo imaging of retinal structures. OCT angiography (OCTA) generates flow information, such as vascular flow, from within the retina. Examples of OCT systems are provided in U.S. Pat. Nos. 6,741,359 and 9,706,915, and examples of OCTA systems are provided in U.S. Pat. Nos. 9,700,206 and 9,759,544, all of which are incorporated herein by reference in their entireties. An exemplary OCT / OCTA system is provided herein.

[0068] FIG. 18 illustrates a generalized frequency-domain optical coherence tomography (FD-OCT) system for collecting 3D image data of the eye suitable for use with the present invention. The FD-OCT system OCT_1 includes a light source LtSrc1. Typical light sources include, but are not limited to, a broadband light source with a short temporal coherence length or a swept laser source. A beam of light from the light source LtSrc1 is typically guided by an optical fiber Fbr1 to illuminate a sample, such as an eye E, a typical sample being human intraocular tissue. The light source LrSrc1 may be, for example, a broadband light source with a short temporal coherence length in the case of spectral-domain OCT (SD-OCT) or a tunable laser source in the case of swept-source OCT (SS-OCT). The light may typically be scanned using a scanner Scnr1 between the output of the optical fiber Fbr1 and the sample E, such that the beam of light (dashed line Bm) is scanned laterally across the region of the sample to be imaged. The light beam from scanner Scnr1 passes through a scan lens SL and an ophthalmic lens OL and can be focused onto the sample E to be imaged. The scan lens SL can receive the light beam from scanner Scnr1 at multiple angles of incidence to generate substantially collimated light, which the ophthalmic lens OL can then focus onto the sample. This example shows a scanning beam that needs to be scanned in two lateral directions (e.g., the x and y directions on a Cartesian plane) to scan a desired field of view (FOV). This example is point-field OCT, which uses a point-field beam to scan across the sample. Thus, scanner Scnr1 is illustratively shown to include two sub-scanners: a first sub-scanner Xscn for scanning the point-field beam across the sample in a first direction (e.g., the horizontal x direction) and a second sub-scanner Yscn for scanning the point-field beam across the sample in an intersecting second direction (e.g., the vertical y direction). If the scanning beam is a line field beam (e.g., line field OCT) and can sample an entire line portion of the sample at a time, only one scanner may be required to scan the line field beam across the sample to span the desired FOV.If the scanning beam is a full-field beam (eg, full-field OCT), a scanner may not be required and the full-field light beam may be illuminated across the entire desired FOV at once.

[0069] Regardless of the type of beam used, light scattered from the sample (e.g., sample light) is collected. In this example, scattered light returning from the sample is collected into the same optical fiber Fbr1 used to route light for illumination. A reference beam derived from the same light source LtSrc1 travels along a separate path, which in this case includes optical fiber Fbr2 and a retroreflector RR1 with an adjustable optical delay. As will be appreciated by those skilled in the art, a transmissive reference path can also be used, with an adjustable delay located in either the sample or the reference arm of the interferometer. The collected sample light is combined with the reference beam, for example, in a fiber coupler Cplr1, to form optical interference within an OCT photodetector Dtctr1 (e.g., a photodetector array, digital camera, etc.). While one fiber port is shown reaching detector Dtctr1, those skilled in the art will appreciate that various interferometer designs can be used for balanced or unbalanced detection of the interference signal. The output from detector Dtctr1 is fed to a processor (e.g., an internal or external computing device) Cmp1, which converts the observed interference into sample depth information. The depth information may be stored in a memory associated with processor Cmp1 and / or displayed on a display (e.g., computer / electronic display / screen) Scn1. The processing and storage functions may be localized within the OCT device, or the functions may be offloaded to (e.g., executed on) an external processor (e.g., an external computer system) to which the collected data is transferred. An example of a computing device (or computer system) is shown in Figure 22. This unit may be dedicated to data processing or may perform other tasks that are quite general and not dedicated to the OCT device.The processor (computing device) Cmp1 may include, for example, a field programmable gate array (FPGA), a digital signal processor (DSP), an application specific integrated circuit (ASIC), a graphics processing unit (GPU), a system on a chip (SoC), a central processing unit (CPU), a general purpose graphics processing unit (GPGPU), or combinations thereof, which may perform some or all of the processing steps in a serial and / or parallel manner with one or more host processors and / or one or more external computing devices.

[0070] The sample and reference arms in an interferometer can be constructed of bulk optics, fiber optics, or hybrid bulk optics systems and can have different architectures, such as Michelson, Mach-Zehnder, or common-path designs, as known to those skilled in the art. Light beam, as used herein, should be interpreted as any carefully directed optical path. Instead of mechanically scanning the beam, a light field can illuminate a one-dimensional or two-dimensional area of ​​the retina to generate OCT data (e.g., U.S. Pat. No. 9,332,902; D. Hillmann et al., "Holoscopy-holographic optical coherence tomography," Optics Letters, Vol. 36(13), p. 2290, 2011; Y. Nakamura et al., "High-Speed ​​three dimensional human retinal imaging by line field spectral domain optical coherence tomography," Optics Express, 2011). Express, 15(12), p. 7103, 2007; Blazkiewicz et al., "Signal-to-noise ratio study of full-field Fourier-domain optical coherence tomography," Applied Optics, 44(36), p. 7722 (2005). In time-domain systems, the reference arm must have an adjustable optical delay to create interference. Balanced detection systems are typically used in TD-OCT and SS-OCT systems, while a spectrometer is used at the detection port for SD-OCT systems. The invention described herein can be applied to either type of OCT system.Various aspects of the present invention may be applied to any type of OCT system or other types of ophthalmic diagnostic systems and / or multiple ophthalmic diagnostic systems, including, but not limited to, fundus imaging systems, visual field testing devices, and scanning laser polarimeters.

[0071] In Fourier-domain optical coherence tomography (FD-OCT), each measurement is a real-valued spectrally controlled interferogram (Sj(k)). Real-valued spectral data typically undergoes several post-processing steps, including background subtraction, dispersion correction, etc. A Fourier transform of the processed interferogram yields a complex OCT signal output Aj(z) = |Aj|eiφ. The absolute value of this complex OCT signal, |Aj|, ​​reveals the scattering intensity at different path lengths and, therefore, the scattering profile with respect to depth (z-direction) within the sample. Similarly, the phase φj can also be extracted from the complex OCT signal. The scattering profile with respect to depth is called an axial scan (A-scan). A collection of A-scans measured at adjacent locations within the sample generates a cross-sectional image (tomogram or B-scan) of the sample. A collection of B-scans collected at different lateral locations on the sample constitutes a data volume or cube. For a particular data volume, the fast axis refers to the scan direction along one B-scan, and the slow axis refers to the axis along which multiple B-scans are collected. The term "cluster scan" may refer to a unit or block of data generated by repeated acquisition at the same (or substantially the same) location (or region) to analyze motion contrast, which may be used to identify blood flow. A cluster scan may consist of multiple A-scans or B-scans collected at approximately the same location on the sample at a relatively short time interval. Because the scans in a cluster scan are of the same region, stationary structures remain relatively unchanged between scans in the cluster scan, while motion contrast between scans that meet predetermined criteria may be identified as blood flow.

[0072] Various methods for generating B-scans are known in the art, including, but not limited to, along the horizontal or x-direction, along the vertical or y-direction, along the x and y diagonals, or in a circular or spiral pattern. B-scans can be in the xz dimension, but can also be any cross-sectional image including the z-dimension. An exemplary OCT B-scan image of a normal retina of a human eye is shown in FIG. 19. An OCT B-scan of the retina provides a view of the structure of the retinal tissue. For illustrative purposes, FIG. 13 identifies the various normal retinal layers and layer boundaries. The identified retinal boundary layers include (from top to bottom) the inner limiting membrane (ILM) layer 1, the retinal nerve fiber layer (RNFL or NFL) layer 2, the ganglion cell layer (GCL) layer 3, the inner plexiform layer (IPL) layer 4, the inner nuclear layer (INL) layer 5, the outer plexiform layer (OPL) layer 6, the outer nuclear layer (ONL) layer 7, the junction between the outer segments (OS) and inner segments (IS) of photoreceptors (indicated by reference numeral 8), the external limiting membrane (ELM or OLM) layer 9, the retinal pigment epithelium (RPE) layer 10, and the Bruch's membrane (BM) layer 11.

[0073] In OCT angiography or functional OCT, analysis algorithms may be applied to OCT data collected at different times (e.g., cluster scans) at the same or nearly the same sample location on the sample to analyze motion or flow (see, e.g., U.S. Patent Application Publication Nos. 2005 / 0171438, 2012 / 0307014, 2010 / 0027857, 2012 / 0277579, and U.S. Patent No. 6,549,801, all of which are incorporated by reference in their entireties). OCT systems may use any one of a number of OCT angiography processing algorithms (e.g., motion contrast algorithms) to identify blood flow. For example, motion contrast algorithms can be applied to intensity information derived from the image data (intensity-based algorithms), phase information from the image data (phase-based algorithms), or complex image data (complex-based algorithms). An en face image is a 2D projection of the 3D OCT data (e.g., by averaging the intensity of each individual A-scan, whereby each A-scan defines a pixel in the 2D projection). Similarly, an en face vascular image is an image that displays motion contrast signals in which the data dimension corresponding to depth (e.g., the z direction along the A-scan) is displayed as a single representative value (e.g., a pixel in the 2D projection image), typically by summing or integrating all or isolated portions of the data (see, e.g., U.S. Pat. No. 7,301,644, incorporated herein by reference in its entirety). OCT systems that provide angiography capabilities may be referred to as OCT angiography (OCTA) systems.

[0074] FIG. 20 shows an example of an en face vasculature image. After processing the data and highlighting motion contrast using any of the motion contrast methods known in the art, an en face (e.g., front view) image of the vasculature may be generated by summing pixel ranges corresponding to a tissue depth from the surface of the retinal internal limiting membrane (ILM). FIG. 21 shows an exemplary B-scan of a vasculature (OCTA) image. As shown, structural information may be less clear because blood flow traverses multiple retinal layers, obscuring them more than in a structural OCT B-scan such as that shown in FIG. 19. Nevertheless, OCTA provides a noninvasive technique for imaging the retinal and choroidal microvasculature, which may be important for diagnosing and / or monitoring various pathologies. For example, OCTA can be used to identify diabetic retinopathy by identifying microaneurysms, neovascular complexes, and quantifying the foveal avascular zone and nonperfused areas. Furthermore, OCTA has been shown to show good agreement with fluorescein angiography (FA), a more traditional but less invasive technique that requires the injection of dye to observe vascular flow in the retina. Furthermore, in dry age-related macular degeneration (AMD), OCTA has been used to monitor the overall decrease in choriocapillaris flow. Similarly, in exudative AMD, OCTA can provide qualitative and quantitative analysis of choroidal neovascular membranes. OCTA has also been used to study vascular obstruction, for example, to assess nonperfused areas and the integrity of the superficial and deep plexuses.

[0075] Computing Devices / Systems FIG. 22 illustrates an exemplary computer system (or computing device). In some embodiments, one or more computer systems may provide functionality described or illustrated herein and / or perform one or more steps of one or more methods described or illustrated herein. The computer system may take any suitable physical form. For example, the computer system may be an embedded computer system, a system-on-chip (SOC), or a single-board computer system (SBC) (e.g., a computer-on-module (COM) or system-on-module (SOM)), a desktop computer system, a laptop or notebook computer system, a mesh of computer systems, a mobile phone, a personal digital assistant (PDA), a server, a tablet computer system, an augmented / virtual reality device, or a combination of two or more of these. Where appropriate, the computer system may reside in the cloud, which may include one or more cloud components within one or more networks.

[0076] In some embodiments, a computer system may include a processor Cpnt1, a memory Cpnt2, a storage Cpnt3, an input / output (I / O) interface Cpnt4, a communication interface Cpnt5, and a bus Cpnt6. The computer system may also optionally include a display Cpnt7, such as a computer monitor or screen.

[0077] The processor Cpnt1 includes hardware for executing instructions, such as those that constitute a computer program. For example, the processor Cpnt1 may be a central processing unit (CPU) or a general-purpose computing-on-graphics processing unit (GPGPU). The processor Cpnt1 may read (or fetch) instructions from an internal register, an internal cache, memory Cpnt2, or storage Cpnt3, decode and execute the instructions, and write one or more results to the internal register, the internal cache, memory Cpnt2, or storage Cpnt3. In particular embodiments, the processor Cpnt1 may include one or more internal caches for data, instructions, or addresses. The processor Cpnt1 may include one or more instruction caches and one or more data caches, for example, to hold data tables. Instructions in the instruction caches may be copies of instructions in memory Cpnt2 or storage Cpnt3, and the instruction caches may speed up retrieval of these instructions by the processor Cpnt1. Processor Cpnt1 may include any suitable number of internal registers and may include one or more arithmetic logic units (ALUs). Processor Cpnt1 may be a multi-core processor or may include one or more processors Cpnt1. Although this disclosure describes and illustrates a particular processor, this disclosure contemplates any suitable processor.

[0078] The memory Cpnt2 may include a main memory that stores instructions for the processor Cpnt1 to execute processing or to hold intermediate data during processing. For example, the computer system may load instructions or data (e.g., a data table) from the storage Cpnt3 or from other sources (e.g., another computer system) into the memory Cpnt2. The processor Cpnt1 may load instructions and data from the memory Cpnt2 into one or more internal registers or internal caches. To execute an instruction, the processor Cpnt1 may read and decode the instruction from the internal register or internal cache. During or after the execution of an instruction, the processor Cpnt1 may write one or more results (which may be intermediate or final results) to an internal register, an internal cache, the memory Cpnt2, or the storage Cpnt3. The bus Cpnt6 may include one or more memory buses (each of which may include an ADDRESS bus and a DATA bus) and may couple the processor Cpnt1 to the memory Cpnt2 and / or the storage Cpnt3. Optionally, one or more memory management units (MMUs) facilitate data transfer between the processor Cpnt1 and the memory Cpnt2. The memory Cpnt2 (which may be a high-speed volatile memory) may include random access memory (RAM), such as dynamic RAM (DRAM) or static RAM (SRAM). The storage Cpnt3 may include long-term or high-capacity storage for data or instructions. The storage Cpnt3 may be internal or external to the computer system and may include one or more of a disk drive (e.g., a hard disk drive (HDD) or a solid-state drive (SSD)), flash memory, ROM, EPROM, optical disk, magneto-optical disk, magnetic tape, a universal serial bus (USB)-accessible drive, or other type of non-volatile memory.

[0079] The I / O interface Cpnt4 may be software, hardware, or a combination of both, and may include one or more interfaces (e.g., serial or parallel communication ports) for communicating with I / O devices, which may enable communication with a human (e.g., a user). For example, the I / O devices may include a keyboard, keypad, microphone, monitor, mouse, printer, scanner, speaker, still camera, stylus, table, touch screen, trackball, video camera, other suitable I / O device, or a combination of two or more thereof.

[0080] The communication interface Cpnt5 may provide a network interface for communicating with other systems or networks. The communication interface Cpnt5 may include a Bluetooth interface or other types of packet-based communication. For example, the communication interface Cpnt5 may include a network interface controller (NIC) and / or a wireless NIC or wireless adapter for communication with a wireless network. The communication interface Cpnt5 may provide communication with a Wi-Fi network, an ad-hoc network, a personal area network (PAN), a wireless PAN (e.g., Bluetooth WPAN), a local area network (LAN), a wide area network (WAN), a metropolitan area network (MAN), a cellular network (e.g., a Global System for Mobile Communications (GSM) network), the Internet, or a combination of two or more thereof.

[0081] Bus Cpnt6 may provide a communication link between the above-mentioned components of the computing system. For example, bus Cpnt6 may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an InfiniBand bus, a low-pin-count (LPC) bus, a memory bus, a MicroChannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCIe) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association local (VLB) bus, or any other suitable bus, or a combination of two or more thereof.

[0082] Although this disclosure describes and illustrates a particular computer system having a particular number of particular components in a particular arrangement, this disclosure contemplates any suitable computer system having any suitable number of any suitable components in any suitable arrangement.

[0083] As used herein, a computer-readable non-transitory storage medium may include one or more semiconductor-based or other integrated circuits (ICs) (e.g., field programmable gate arrays (FPGAs) or application-specific ICs (ASICs)), hard disk drives (HDDs), hybrid hard drives (HHDs), optical disks, optical disk drives (ODDs), magneto-optical disks, magneto-optical drives, floppy diskettes, floppy disk drives (FDDs), magnetic tapes, solid-state drives (SSDs), RAM-drives, SECURE DIGITAL cards or drives, or any other suitable computer-readable non-transitory storage medium, or any suitable combination of two or more thereof, where appropriate. A computer-readable non-transitory storage medium may be volatile, non-volatile, or a combination of volatile and non-volatile, where appropriate.

[0084] While the present invention has been described in conjunction with several specific embodiments, as will be apparent to those skilled in the art in light of the foregoing description, many other alternatives, modifications, and variations will be apparent. Accordingly, the invention as described herein is intended to embrace all such alternatives, modifications, applications, and variations that may fall within the spirit and scope of the appended claims.

Claims

1. 1. A method for segmenting retinal layers in an optical coherence tomography (OCT) image, comprising: collecting OCT data of the eye using an OCT system; segmenting the OCT data into individual retinal layers; For the target retinal layer, creating a first en face image based on a first slab, wherein the location of the target retinal layer is near a bottom of the first slab; creating a second en face image based on a second slab, wherein the target retinal layer is located between the top layer and the bottom layer of the second slab; and designating the segmentation of the target retinal layer as unsuccessful or successful based on a similarity measure between the first en face image and the second en face image.

2. The method of claim 1 , wherein the first en face image and the second en face image are of slabs from the OCT data or from OCT angiography data generated using the OCT data.

3. The method of claim 1 or 2, wherein the similarity measure is based on normalized cross-correlation (NCC) between en face images.

4. 4. The method of claim 1, wherein in response to a segmentation being designated as unsuccessful, at least a portion of the segmentation is replaced with an approximation based on the top layer and the bottom layer of the second slab.

5. 5. The method of claim 4, wherein determining the approximation includes applying weights to the top and bottom layers of the second slab based on their positions relative to an expected position of the target retinal layer.

6. The method of claim 4 , wherein the similarity measure includes a local similarity measure that identifies local segmentation failures, the local segmentation failures being replaced by the approximation values.

7. The method of claim 6 , wherein the local similarity measure is determined for each B-scan.

8. The method of claim 1 , wherein the first slab and the second slab have the same top layer.

9. The method of claim 8 , wherein the top layer and the bottom layer of the second slab are selected based on a light-to-dark or dark-to-light sharpness transition measure.

10. The method of claim 8 , wherein the target retinal layer is the lowest layer of the first slab and the lowest layer of the second slab is below the target retinal layer.

11. 11. The method of claim 10, wherein the uppermost layer is the internal limiting membrane (ILM), the target retinal layer is the inner plexiform layer (IPL), and the lowermost layer of the second slab is the outer plexiform layer (OPL).

12. 11. The method of claim 10, wherein the uppermost layer is the internal limiting membrane (ILM), the target retinal layer is the outer plexiform layer (OPL), and the lowermost layer of the second slab is the junction between the outer segment (OS) and the inner segment (IS).

Images