Retinal layer segmentation in optical coherence tomography (OCT) using a 3D deep-convolutional regression network for patients with age-related macular degeneration

Introduction – Retinal layer segmentation in optical coherence tomography (OCT) images is an important approach for detecting and prognosing disease. Automating segmentation using robust machine learning techniques lead to computationally efficient solutions and significantly reduces the cost of labor-intensive labeling, which is traditionally performed by trained graders at a reading center, sometimes aided by semi-automated algorithms. Although several algorithms have been proposed since the revival of deep learning, eyes with severe pathological conditions continue to challenge fully automated segmentation approaches. There remains an opportunity to leverage the underlying spatial correlations between the retinal surfaces in the segmentation approach. Methods - Some of these proposed traditional methods can be expanded to utilize the three-dimensional spatial context governing the retinal image volumes by replacing the use of 2D filters with 3D filters. Towards this purpose, we propose a spatial-context, continuity and anatomical relationship preserving semantic segmentation algorithm, which utilizes the 3D spatial context from the image volumes with the use of 3D filters. We propose a 3D deep neural network capable of learning the surface positions of the layers in the retinal volumes. Results - We utilize a dataset of OCT images from patients with Age-related Macular Degeneration (AMD) to assess performance of our model and provide both qualitative (including segmentation maps and thickness maps) and quantitative (including error metric comparisons and volumetric comparisons) results, which demonstrate that our proposed method performs favorably even for eyes with pathological changes caused by severe retinal diseases. The Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE) for patients with a wide range of AMD severity scores (0–11) were within 0.84±0.41 and 1.33±0.73 pixels, respectively, which are significantly better than some of the other state-of-the-art algorithms. Conclusion – The results demonstrate the utility of extracting features from the entire OCT volume by treating the volume as a correlated entity and show the benefit of utilizing 3D autoencoder based regression networks for smoothing the approximated retinal layers by inducing shape based regularization constraints.     

Human retinal imaging using visible-light optical coherence tomography guided by scanning laser ophthalmoscopy

We achieved human retinal imaging using visible-light optical coherence tomography (vis-OCT) guided by an integrated scanning laser ophthalmoscopy (SLO). We adapted a spectral domain OCT configuration and used a supercontinuum laser as the illumating source. The center wavelength was 564 nm and the bandwidth was 115 nm, which provided a 0.97 µm axial resolution measured in air. We characterized the sensitivity to be 86 dB with 226 µW incidence power on the pupil. We also integrated an SLO that shared the same optical path of the vis-OCT sample arm for alignment purposes. We demonstrated the retinal imaging from both systems centered at the fovea and optic nerve head with 20° × 20° and 10° × 10° field of view. We observed similar anatomical structures in vis-OCT and NIR-OCT. The contrast appeared different from vis-OCT to NIR-OCT, including slightly weaker signal from intra-retinal layers, and increased visibility and contrast of anatomical layers in the outer retina.OCIS codes: (110.4190) Multiple imaging, (170.0110) Imaging systems, (170.4470) Ophthalmology, (170.4500) Optical coherence tomography     

Feasibility of the soft attention-based models for automatic segmentation of OCT kidney images

Clinically, optical coherence tomography (OCT) has been utilized to obtain the images of the kidney’s proximal convoluted tubules (PCTs), which can be used to quantify the morphometric parameters such as tubular density and diameter. Such parameters are useful for evaluating the status of the donor kidney for transplant. Quantifying PCTs from OCT images by human readers is a time-consuming and tedious process. Despite the fact that conventional deep learning models such as conventional neural networks (CNNs) have achieved great success in the automatic segmentation of kidney OCT images, gaps remain regarding the segmentation accuracy and reliability. Attention-based deep learning model has benefits over regular CNNs as it is intended to focus on the relevant part of the image and extract features for those regions. This paper aims at developing an Attention-based UNET model for automatic image analysis, pattern recognition, and segmentation of kidney OCT images. We evaluated five methods including the Residual-Attention-UNET, Attention-UNET, standard UNET, Residual UNET, and fully convolutional neural network using 14403 OCT images from 169 transplant kidneys for training and testing. Our results show that Residual-Attention-UNET outperformed the other four methods in segmentation by showing the highest values of all the six metrics including dice score (0.81 ± 0.01), intersection over union (IOU, 0.83 ± 0.02), specificity (0.84 ± 0.02), recall (0.82 ± 0.03), precision (0.81 ± 0.01), and accuracy (0.98 ± 0.08). Our results also show that the performance of the Residual-Attention-UNET is equivalent to the human manual segmentation (dice score = 0.84 ± 0.05). Residual-Attention-UNET and Attention-UNET also demonstrated good performance when trained on a small dataset (3456 images) whereas the performance of the other three methods dropped dramatically. In conclusion, our results suggested that the soft Attention-based models and specifically Residual-Attention-UNET are powerful and reliable methods for tubule lumen identification and segmentation and can help clinical evaluation of transplant kidney viability as fast and accurate as possible.     

Framework for quantitative three-dimensional choroidal vasculature analysis using optical coherence tomography

Choroidal vasculature plays an important role in the pathogenesis of retinal diseases, such as myopic maculopathy, age-related macular degeneration, diabetic retinopathy, central serous chorioretinopathy, and ocular inflammatory diseases. Current optical coherence tomography (OCT) technology provides three-dimensional visualization of the choroidal angioarchitecture; however, quantitative measures remain challenging. Here, we propose and validate a framework to segment and quantify the choroidal vasculature from a prototype swept-source OCT (PLEX Elite 9000, Carl Zeiss Meditec, USA) using a 3×3 mm scan protocol centered on the macula. Enface images referenced from the retinal pigment epithelium were reconstructed from the volumetric data. The boundaries of the choroidal volume were automatically identified by tracking the choroidal vessel feature structure over the depth, and a selective sliding window was applied for segmenting the vessels adaptively from attenuation-corrected enface images. We achieved a segmentation accuracy of 96% ± 1% as compared with manual annotation, and a dice coefficient of 0.83 ± 0.04 for repeatability. Using this framework on both control (0.00 D to −2.00 D) and highly myopic (−8.00 D to −11.00 D) eyes, we report a decrease in choroidal vessel volume (p<0.001) in eyes with high myopia.     

OCT-GAN: single step shadow and noise removal from optical coherence tomography images of the human optic nerve head

Speckle noise and retinal shadows within OCT B-scans occlude important edges, fine textures and deep tissues, preventing accurate and robust diagnosis by algorithms and clinicians. We developed a single process that successfully removed both noise and retinal shadows from unseen single-frame B-scans within 10.4ms. Mean average gradient magnitude (AGM) for the proposed algorithm was 57.2% higher than current state-of-the-art, while mean peak signal to noise ratio (PSNR), contrast to noise ratio (CNR), and structural similarity index metric (SSIM) increased by 11.1%, 154% and 187% respectively compared to single-frame B-scans. Mean intralayer contrast (ILC) improvement for the retinal nerve fiber layer (RNFL), photoreceptor layer (PR) and retinal pigment epithelium (RPE) layers decreased from 0.362 ± 0.133 to 0.142 ± 0.102, 0.449 ± 0.116 to 0.0904 ± 0.0769, 0.381 ± 0.100 to 0.0590 ± 0.0451 respectively. The proposed algorithm reduces the necessity for long image acquisition times, minimizes expensive hardware requirements and reduces motion artifacts in OCT images.     

Imaging of retinal vasculature using adaptive optics SLO/OCT

We use our previously developed adaptive optics (AO) scanning laser ophthalmoscope (SLO)/ optical coherence tomography (OCT) instrument to investigate its capability for imaging retinal vasculature. The system records SLO and OCT images simultaneously with a pixel to pixel correspondence which allows a direct comparison between those imaging modalities. Different field of views ranging from 0.8°x0.8° up to 4°x4° are supported by the instrument. In addition a dynamic focus scheme was developed for the AO-SLO/OCT system in order to maintain the high transverse resolution throughout imaging depth. The active axial eye tracking that is implemented in the OCT channel allows time resolved measurements of the retinal vasculature in the en-face imaging plane. Vessel walls and structures that we believe correspond to individual erythrocytes could be visualized with the system.OCIS codes: (170.3890) Medical optics instrumentation, (110.1080) Active or adaptive optics, (170.4470) Ophthalmology, (330.5310) Vision - photoreceptors, (110.4500) Optical coherence tomography     

Ultra-widefield retinal MHz-OCT imaging with up to 100 degrees viewing angle

We evaluate strategies to maximize the field of view (FOV) of in vivo retinal OCT imaging of human eyes. Three imaging modes are tested: Single volume imaging with 85° FOV as well as with 100° and stitching of five 60° images to a 100° mosaic (measured from the nodal point). We employ a MHz-OCT system based on a 1060nm Fourier domain mode locked (FDML) laser with a depth scan rate of 1.68MHz. The high speed is essential for dense isotropic sampling of the large areas. Challenges caused by the wide FOV are discussed and solutions to most issues are presented. Detailed information on the design and characterization of our sample arm optics is given. We investigate the origin of an angle dependent signal fall-off which we observe towards larger imaging angles. It is present in our 85° and 100° single volume images, but not in the mosaic. Our results suggest that 100° FOV OCT is possible with current swept source OCT technology.OCIS codes: (170.4500) Optical coherence tomography, (170.3880) Medical and biological imaging, (170.4460) Ophthalmic optics and devices, (120.3890) Medical optics instrumentation, (140.3510) Lasers, fiber     

In vivo imaging of retinal hemodynamics with OCT angiography and Doppler OCT

Retinal hemodynamics is important for early diagnosis and precise monitoring in retinal vascular diseases. We propose a novel method for measuring absolute retinal blood flow in vivo using the combined techniques of optical coherence tomography (OCT) angiography and Doppler OCT. Doppler values can be corrected by Doppler angles extracted from OCT angiography images. A three-dimensional (3D) segmentation algorithm based on dynamic programming was developed to extract the 3D boundaries of optic disc vessels, and Doppler angles were calculated from 3D vessel geometry. The accuracy of blood flow from the Doppler OCT was validated using a flow phantom. The feasibility of the method was tested on a subject in vivo. The pulsatile retinal blood flow and the parameters for retinal hemodynamics were successfully obtained.OCIS codes: (100.2960) Image analysis, (100.6890) Three-dimensional image processing, (170.4500) Optical coherence tomography, (170.4460) Ophthalmic optics and devices     

MF-AV-Net: an open-source deep learning network with multimodal fusion options for artery-vein segmentation in OCT angiography

This study is to demonstrate the effect of multimodal fusion on the performance of deep learning artery-vein (AV) segmentation in optical coherence tomography (OCT) and OCT angiography (OCTA); and to explore OCT/OCTA characteristics used in the deep learning AV segmentation. We quantitatively evaluated multimodal architectures with early and late OCT-OCTA fusions, compared to the unimodal architectures with OCT-only and OCTA-only inputs. The OCTA-only architecture, early OCT-OCTA fusion architecture, and late OCT-OCTA fusion architecture yielded competitive performances. For the 6 mm×6 mm and 3 mm×3 mm datasets, the late fusion architecture achieved an overall accuracy of 96.02% and 94.00%, slightly better than the OCTA-only architecture which achieved an overall accuracy of 95.76% and 93.79%. 6 mm×6 mm OCTA images show AV information at pre-capillary level structure, while 3 mm×3 mm OCTA images reveal AV information at capillary level detail. In order to interpret the deep learning performance, saliency maps were produced to identify OCT/OCTA image characteristics for AV segmentation. Comparative OCT and OCTA saliency maps support the capillary-free zone as one of the possible features for AV segmentation in OCTA. The deep learning network MF-AV-Net used in this study is available on GitHub for open access.     

Structural and functional retinal changes in patients with type 2 diabetes without diabetic retinopathy

ObjectiveThe characteristics of the early changes in preclinical diabetic retinopathy (DR) are poorly known. This study aimed to analyse the changes in the structure and function of the fundus in diabetic patients without diabetic retinopathy (NDR).MethodsThis prospective study enrolled patients with type 2 diabetes and healthy controls from April to December 2020. Retinal sensitivity was measured by microperimetry. The peripapillary retinal nerve fibre layer (p-RNFL) thickness, macular retinal thickness, and retinal volume were measured by optical coherence tomography (OCT). The vessel density (VD) and perfusion density (PD) of the peripapillary area, as well as the foveal avascular zone (FAZ) area, FAZ perimeter, and FAZ circularity, were measured by optical coherence tomographic angiography (OCTA).ResultsA total of 71 cases (100 eyes) were enrolled in the study, including 34 cases (51 eyes) in the NDR group and 37 cases (49 eyes) in the control group. The mean retinal sensitivity was lower in the NDR group than in the control group for all sectors (all p < .001). Compared with controls, the NDR group showed thinner p-RNFL in the T sector (76.24 ± 14.29 vs. 85.47 ± 19.66 µm, p = .035). The NDR group had a thinner retina in the N2 sector (304.55 ± 16.07 vs. 312.02 ± 12.30 µm, p = .010). The PD of DCP was lower in the N2 sector in the NDR group (44.92 ± 11.77 vs. 50.27 ± 6.37%, p = .044). The VD was higher in the NDR group in RPCP-S/N/I, and the PD was higher in the RPCP-S/N (all p < .05). The frequencies of perifoveal capillary drop-out, notched or punched out borders of the superficial FAZ, and loss of smooth contour were all higher in the NDR group (all p < .05).ConclusionThe structure (p-RNFL thickness, VD, and PD) and function (retinal sensitivity) display some changes in diabetic patients even if they had not been found to have DR.

Key messages

1. Decreased retinal sensitivity was observed in diabetic patients before the onset of diabetic retinopathy.
2. Compared with the control group, we found the changes in vessel density or perfusion density in a certain area, whether in SCP, DCP, or RPCP in the NDR group.
3. Before the onset of diabetic retinopathy, the structure and function of the retina in diabetic patients had changed.

     

Developing a normative database for retinal perfusion using optical coherence tomography angiography

Visualizing and characterizing microvascular abnormalities with optical coherence tomography angiography (OCTA) has deepened our understanding of ocular diseases, such as glaucoma, diabetic retinopathy, and age-related macular degeneration. Two types of microvascular defects can be detected by OCTA: focal decrease because of localized absence and collapse of retinal capillaries, which is referred to as the non-perfusion area in OCTA, and diffuse perfusion decrease usually detected by comparing with healthy case-control groups. Wider OCTA allows for insights into peripheral retinal vascularity, but the heterogeneous perfusion distribution from the macula, parapapillary area to periphery hurdles the quantitative assessment. A normative database for OCTA could estimate how much individual’s data deviate from the normal range, and where the deviations locate. Here, we acquired OCTA images using a swept-source OCT system and a 12×12 mm protocol in healthy subjects. We automatically segmented the large blood vessels with U-Net, corrected for anatomical factors such as the relative position of fovea and disc, and segmented the capillaries by a moving window scheme. A total of 195 eyes were included and divided into 4 age groups: < 30 (n=24) years old, 30-49 (n=28) years old, 50-69 (n=109) years old and >69 (n=34) years old. This provides an age-dependent normative database for characterizing retinal perfusion abnormalities in 12×12 mm OCTA images. The usefulness of the normative database was tested on two pathological groups: one with diabetic retinopathy; the other with glaucoma.     

Retinal optical coherence tomography image analysis by a restricted Boltzmann machine

Optical coherence tomography (OCT) is an emerging imaging technique for ophthalmic disease diagnosis. Two major problems in OCT image analysis are image enhancement and image segmentation. Deep learning methods have achieved excellent performance in image analysis. However, most of the deep learning-based image analysis models are supervised learning-based approaches and need a high volume of training data (e.g., reference clean images for image enhancement and accurate annotated images for segmentation). Moreover, acquiring reference clean images for OCT image enhancement and accurate annotation of the high volume of OCT images for segmentation is hard. So, it is difficult to extend these deep learning methods to the OCT image analysis. We propose an unsupervised learning-based approach for OCT image enhancement and abnormality segmentation, where the model can be trained without reference images. The image is reconstructed by Restricted Boltzmann Machine (RBM) by defining a target function and minimizing it. For OCT image enhancement, each image is independently learned by the RBM network and is eventually reconstructed. In the reconstruction phase, we use the ReLu function instead of the Sigmoid function. Reconstruction of images given by the RBM network leads to improved image contrast in comparison to other competitive methods in terms of contrast to noise ratio (CNR). For anomaly detection, hyper-reflective foci (HF) as one of the first signs in retinal OCTs of patients with diabetic macular edema (DME) are identified based on image reconstruction by RBM and post-processing by removing the HFs candidates outside the area between the first and the last retinal layers. Our anomaly detection method achieves a high ability to detect abnormalities.     

Handheld simultaneous scanning laser ophthalmoscopy and optical coherence tomography system

Scanning laser ophthalmoscopy (SLO) and optical coherence tomography (OCT) are widely used retinal imaging modalities that can assist in the diagnosis of retinal pathologies. The combination of SLO and OCT provides a more comprehensive imaging system and a method to register OCT images to produce motion corrected retinal volumes. While high quality, bench-top SLO-OCT systems have been discussed in the literature and are available commercially, there are currently no handheld designs. We describe the first design and fabrication of a handheld SLO/spectral domain OCT probe. SLO and OCT images were acquired simultaneously with a combined power under the ANSI limit. High signal-to-noise ratio SLO and OCT images were acquired simultaneously from a normal subject with visible motion artifacts. Fully automated motion estimation methods were performed in post-processing to correct for the inter- and intra-frame motion in SLO images and their concurrently acquired OCT volumes. The resulting set of reconstructed SLO images and the OCT volume were without visible motion artifacts. At a reduced field of view, the SLO resolved parafoveal cones without adaptive optics at a retinal eccentricity of 11° in subjects with good ocular optics. This system may be especially useful for imaging young children and subjects with less stable fixation.OCIS codes: (170.4460) Ophthalmic optics and devices, (080.3620) Lens system design, (170.0110) Imaging systems, (170.5755) Retina scanning, (170.4470) Ophthalmology, (110.4500) Optical coherence tomography, (110.4153) Motion estimation and optical flow     

PCA-based localization approach for segmentation of optic disc

Purpose

The optic disc is the origin of the optic nerve, where the axons of retinal ganglion cells join together. The size, shape and contour of optic disc are used for classification and identification of retinal diseases. Automatic detection of eye disease requires development of an efficient algorithm. This paper proposes an efficient method for optic disc segmentation and detection for the diagnosis of retinal diseases.

Methods

The methodology involves optic disc localization, blood vessel inpainting and optic disc segmentation. Localization is based on principal component analysis, and segmentation is based on Markov random field segmentation. In order to get reasonable background images, blood vessel inpainting is done before segmentation.

Results

The proposed method tested with two standard databases MESSIDOR and DRIVE, and achieved an average overlapping score of 92.41, 92.17%, respectively; also validation experiments were done with one local database from Venu Eye Hospital, New Delhi, and obtained an average overlapping score of 91%.

Conclusion

An efficient algorithm is developed for detecting optic disc using principal component analysis-based localization and Markov random field segmentation. The comparison with alternative method yielded results that demonstrate the superiority of the proposed algorithm for optic disc detection.

     

Fast detection of the optic disc and fovea in color fundus photographs

A fully automated, fast method to detect the fovea and the optic disc in digital color photographs of the retina is presented. The method makes few assumptions about the location of both structures in the image. We define the problem of localizing structures in a retinal image as a regression problem. A kNN regressor is utilized to predict the distance in pixels in the image to the object of interest at any given location in the image based on a set of features measured at that location. The method combines cues measured directly in the image with cues derived from a segmentation of the retinal vasculature. A distance prediction is made for a limited number of image locations and the point with the lowest predicted distance to the optic disc is selected as the optic disc center. Based on this location the search area for the fovea is defined. The location with the lowest predicted distance to the fovea within the foveal search area is selected as the fovea location. The method is trained with 500 images for which the optic disc and fovea locations are known. An extensive evaluation was done on 500 images from a diabetic retinopathy screening program and 100 specially selected images containing gross abnormalities. The method found the optic disc in 99.4% and the fovea in 96.8% of regular screening images and for the images with abnormalities these numbers were 93.0% and 89.0% respectively.     

Relationship between peripapillary choroidal thickness and retinal nerve fiber layer in young people with myopia

AimTo study the characteristics and relationship between peripapillary retinal nerve fiber layer (RNFL) and choroidal thickness in young people with myopia.MethodsWe retrospectively analyzed 92 cases (52 myopia, 40 emmetropia) regarding age, sex, refractive power, axial length (AL), and intraocular pressure. Peripapillary RNFL and choroidal thicknesses were measured by optical coherence tomography (OCT) in six sectors. Differences in thicknesses between the two groups were compared by single-factor analysis.ResultsRNFL was thickest in the inferotemporal sector (157.3 ± 19.66 µm) and thinnest in the nasal sector (58.78 ± 18.41 µm). Peripapillary choroid was thickest in the superonasal sector (176.37 ± 33.92 µm) and thinnest in the inferotemporal sector (131.79 ± 25.22 µm). The RNFL was thinner in the myopia group (99.04 ± 8.23 µm) vs the emmetropia group (103.25 ± 8.32 µm); significantly different in the superotemporal and inferonasal sectors. Peripapillary choroid thickness in the myopia group (148.65 ± 26.64 µm) was lower vs the emmetropia group (160.88 ± 29.06 µm); significantly different in the nasal, inferonasal, and inferotemporal sectors. RNFL thickness was negatively correlated with choroidal thickness in the nasal sector (r = −0.288).ConclusionPeripapillary RNFL and choroidal thicknesses showed regional distributions. RNFL was negatively correlated with PCT in the nasal sector, possibly related to eye axis growth and choroidal compensation.     

光学相干断层扫描观察2型糖尿病患者视盘周围视网膜神经纤维层厚度及黄斑区视网膜厚度的变化

【目的】应用光学相干断层扫描（OCT）观察2型糖尿病患者视盘周围视网膜神经纤维层（RNFL）厚度及黄斑区视网膜厚度的变化。【方法】采用OCT测量正常对照组、无糖尿病视网膜病变（NDR）组及非增殖期糖尿病视网膜病变（NPDR）组患者视盘周围RNFL厚度及黄斑区视网膜厚度的变化。【结果】在视乳头旁RNFL厚度比较中,NDR组与正常对照组比较变薄,NPDR组与正常对照组比较变厚。在黄斑区视网膜厚度比较中, NDR组与正常对照组比较变薄,NPDR组与正常对照组比较变厚。【结论】OCT能定量观察糖尿病患者视盘周围RNFL厚度及黄斑区视网膜厚度的变化,为糖尿病视网膜病变的诊断及治疗提供可靠的检测手段。     

A combined machine-learning and graph-based framework for the segmentation of retinal surfaces in SD-OCT volumes

Optical coherence tomography is routinely used clinically for the detection and management of ocular diseases as well as in research where the studies may involve animals. This routine use requires that the developed automated segmentation methods not only be accurate and reliable, but also be adaptable to meet new requirements. We have previously proposed the use of a graph-theoretic approach for the automated 3-D segmentation of multiple retinal surfaces in volumetric human SD-OCT scans. The method ensures the global optimality of the set of surfaces with respect to a cost function. Cost functions have thus far been typically designed by hand by domain experts. This difficult and time-consuming task significantly impacts the adaptability of these methods to new models. Here, we describe a framework for the automated machine-learning based design of the cost function utilized by this graph-theoretic method. The impact of the learned components on the final segmentation accuracy are statistically assessed in order to tailor the method to specific applications. This adaptability is demonstrated by utilizing the method to segment seven, ten and five retinal surfaces from SD-OCT scans obtained from humans, mice and canines, respectively. The overall unsigned border position errors observed when using the recommended configuration of the graph-theoretic method was 6.45 ± 1.87 μm, 3.35 ± 0.62 μm and 9.75 ± 3.18 μm for the human, mouse and canine set of images, respectively.OCIS codes: (100.0100) Image processing, (100.2000) Digital image processing, (100.4994) Pattern recognition, image transforms, (100.6890) Three-dimensional image processing, (110.4500) Optical coherence tomography     

Adaptive optics optical coherence tomography at 1 MHz

Image acquisition speed of optical coherence tomography (OCT) remains a fundamental barrier that limits its scientific and clinical utility. Here we demonstrate a novel multi-camera adaptive optics (AO-)OCT system for ophthalmologic use that operates at 1 million A-lines/s at a wavelength of 790 nm with 5.3 μm axial resolution in retinal tissue. Central to the spectral-domain design is a novel detection channel based on four high-speed spectrometers that receive light sequentially from a 1 × 4 optical switch assembly. Absence of moving parts enables ultra-fast (50ns) and precise switching with low insertion loss (−0.18 dB per channel). This manner of control makes use of all available light in the detection channel and avoids camera dead-time, both critical for imaging at high speeds. Additional benefit in signal-to-noise accrues from the larger numerical aperture afforded by the use of AO and yields retinal images of comparable dynamic range to that of clinical OCT. We validated system performance by a series of experiments that included imaging in both model and human eyes. We demonstrated the performance of our MHz AO-OCT system to capture detailed images of individual retinal nerve fiber bundles and cone photoreceptors. This is the fastest ophthalmic OCT system we know of in the 700 to 915 nm spectral band.OCIS codes: (110.1080) Active or adaptive optics, (170.4500) Optical coherence tomography, (120.3890) Medical optics instrumentation, (170.0110) Imaging systems, (170.4470) Ophthalmology, (330.5310) Vision - photoreceptors     

Automatic geographic atrophy segmentation using optical attenuation in OCT scans with deep learning

A deep learning algorithm was developed to automatically identify, segment, and quantify geographic atrophy (GA) based on optical attenuation coefficients (OACs) calculated from optical coherence tomography (OCT) datasets. Normal eyes and eyes with GA secondary to age-related macular degeneration were imaged with swept-source OCT using 6 × 6 mm scanning patterns. OACs calculated from OCT scans were used to generate customized composite en face OAC images. GA lesions were identified and measured using customized en face sub-retinal pigment epithelium (subRPE) OCT images. Two deep learning models with the same U-Net architecture were trained using OAC images and subRPE OCT images. Model performance was evaluated using DICE similarity coefficients (DSCs). The GA areas were calculated and compared with manual segmentations using Pearson’s correlation and Bland-Altman plots. In total, 80 GA eyes and 60 normal eyes were included in this study, out of which, 16 GA eyes and 12 normal eyes were used to test the models. Both models identified GA with 100% sensitivity and specificity on the subject level. With the GA eyes, the model trained with OAC images achieved significantly higher DSCs, stronger correlation to manual results and smaller mean bias than the model trained with subRPE OCT images (0.940 ± 0.032 vs 0.889 ± 0.056, p = 0.03, paired t-test, r = 0.995 vs r = 0.959, mean bias = 0.011 mm vs mean bias = 0.117 mm). In summary, the proposed deep learning model using composite OAC images effectively and accurately identified, segmented, and quantified GA using OCT scans.