RetinalFRNet: retinal vessel segmentation in OCTA images using the feature reconstruction network.

Tears of the retinal pigment epithelium (RPE) usually cause a deleterious loss in visual acuity and are most commonly associated with previous vascularized retinal pigment epithelium detachment (vPED) due to age-related macular degeneration (AMD) (Pauleikhoff et al. 2002). Recent studies suggest an increase in RPE tear incidences since the introduction of anti-vascular endothelial growth factor (VEGF) therapies. Therefore, clinicians increasingly direct the focus on the challenge of RPE tear prevention. Several prognostic markers for an impending RPE tear have been described such as vPED lesion's height, hyper-reflective lines in near-infrared images, subretinal clefts, microrips and duration of vPED (Clemens & Eter 2016). Chan and co-workers (2007) firstly described the morphologic proportions of choroidal neovascularization (CNV) and PED lesion as particularly relevant for the likelihood of tear development. A small ratio of CNV size to PED size turned out to correlate with a higher rate of RPE tears. This concept of CNV/PED ratio was based on fluorescence angiographic (FA) data. We report the applicability of the CNV/PED ratio to optical coherence tomography angiography (OCT-A) imaging. Two patients underwent spectral-domain optical coherence tomography and FA imaging as well as OCT-A examinations (Fig. 1). The first patient showed a serous-vascularized PED due to AMD. Fluorescence angiography (FA) shows a shading background fluorescence in the area of the PED in the early phase, followed by an irregular hyperfluorescence at the margin of the PED corresponding to the CNV, which results in a CNV/PED ratio of approximately 0.2 (Fig. 1A). Correspondingly, OCT-A reveals a distinct flow signal within a neovascular network underneath the PED lesion. The entire network is contained in the detached area resulting in a comparable ratio of 0.3 (Fig. 1C). In the second patient, FA imaging revealed a fibrovascular PED with underlying occult CNV and a CNV/PED ratio of approximately 0.7 (Fig. 1B). Corresponding OCT-A shows a neovascular network almost entirely filling out the PED lesion resulting in a CNV/PED ratio of 0.8 (Fig. 1D). OCT-A imaging produces three-dimensional data. Therefore, measurements of vPED lesions can be performed in different retinal layers. Accordingly, vPED and CNV dimensions may vary in OCT-A. Two-dimensional FA images usually visualize the largest lesion diameter. Accordingly, OCT-A measurements were performed in the layer in which vPED and CNV dimensions showed their maximum diameters. Notably, projection artefacts from retinal vessels may influence the visualization of sub-RPE CNV in OCT-A imaging, which must be taken into consideration when determining CNV dimensions (Spaide et al. 2015). OCT-A imaging may be a useful adjunct to the conventional method of FA for CNV/PED ratio assessment as OCT-A visualizes the actual CNV dimensions, whereas CNV measurement in FA is significantly confounded by dye leakage and usually remains an estimate. A critical issue in OCT-A imaging of CNVs under PED represents the slab thickness. A thin slab may not include the maximal dimensions of the CNV membrane resulting in a false CNV/PED ratio, while a thick slab is prone to produce image artefacts. In conclusion, OCT-A allows for a non-invasive assessment of the CNV/PED ratio in vPED patients that represents an important predictive factor of RPE tear development at the beginning as well as in the course of an anti-VEGF treatment. A future study including a higher number of patients must evaluate in how far dimensions of CNV membrane under the PED agree between FA and OCT-A, which will be crucial regarding the relevance as well as the clinical interpretation of the CNV/PED ratio assessed by OCT-A as a risk factor for impending RPE tear development.

To compare the vascular parameters derived from optical coherence tomography angiography (OCTA) and fundus fluorescein angiography (FFA) images. Twenty-two eyes of 22 patients were imaged with OCTA and FFA. FFA images were cropped to the same dimension as OCTA images after registration. Vessel enhancement using a Frangi filter and local fractal analysis was applied to the superficial layer of the OCTA and cropped FFA images. Foveal avascular zone (FAZ) area, vessel density, spacing between large vessels, and spacing between small vessels were quantified. FAZ area was similar between original OCTA, Frangi-filtered OCTA, and FFA images (P = .32). Actual OCTA images had significantly higher vessel density (35.2% ± 1.45%, P < .001) than Frangi-filtered OCTA images (29.8% ± 0.78%) and Frangi-filtered FFA images (25.5% ± 2.41%). Spacing between large vessels was significantly lower in original OCTA images (31.9% ± 1.47%, P < .001) than Frangi-filtered OCTA images (36.8% ± 1.24%) and Frangi-filtered FFA images (60.1% ± 2.68%). Further, FFA images had significantly lower spacing between small vessels (14.4% ± 0.43%, P < .001) than original OCTA images (32.7% ± 1.03%) and Frangi-filtered OCTA images (33.4% ± 0.81%). FAZ area was similar between OCTA and FFA, independent of vessel enhancement. However, vessel enhancement improved the agreement of vascular parameters between OCTA and FFA images of the same eye. [Ophthalmic Surg Lasers Imaging Retina. 2018;49:e175-e181.].

Automatic blood vessels segmentation based on different retinal maps from OCTA scans

The decorrelation signals in optical coherence tomography angiography (OCTA) are derived from the flow of erythrocytes and concomitantly delineate the retinal vasculature. We compared the structural and functional characteristics of vascular lesions visualized in fluorescein angiography (FA), OCTA, and en-face OCT images in 53 eyes (28 patients) with diabetic retinopathy (DR). The foveal avascular zone (FAZ) areas in OCTA images in the superficial layer almost corresponded to those in FA images. The FAZ areas in the en-face OCT images in the superficial layer were smaller than those in the FA images and correlated with each other, which agreed with the finding that en-face OCT images often delineated the vascular structure in the nonperfused areas in FA images. Microaneurysms appeared as fusiform, saccular, or coiled capillaries in OCTA images and ringed, round, or oval hyperreflective lesions in en-face OCT images. OCTA and en-face OCT images detected 41.0 ± 16.1% and 40.1 ± 18.6%, respectively, of microaneurysms in FA images, although both depicted only 13.9 ± 16.4%. The number of microaneurysms in FA images was correlated with that in OCTA and en-face OCT images. Comparisons of these modalities showed the associations and dissociations between blood flow and vascular structures, which improves the understanding of the pathogenesis of DR.

We present an image projection network (IPN), which is a novel end-to-end architecture and can achieve 3D-to-2D image segmentation in optical coherence tomography angiography (OCTA) images. Our key insight is to build a projection learning module (PLM) which uses a unidirectional pooling layer to conduct effective features selection and dimension reduction concurrently. By combining multiple PLMs, the proposed network can input 3D OCTA data, and output 2D segmentation results such as retinal vessel segmentation. It provides a new idea for the quantification of retinal indicators: without retinal layer segmentation and without projection maps. We tested the performance of our network for two crucial retinal image segmentation issues: retinal vessel (RV) segmentation and foveal avascular zone (FAZ) segmentation. The experimental results on 316 OCTA volumes demonstrate that the IPN is an effective implementation of 3D-to-2D segmentation networks, and the uses of multi-modality information and volumetric information make IPN perform better than the baseline methods.

The purpose of this study is to evaluate layer fusion options for deep learning classification of optical coherence tomography (OCT) angiography (OCTA) images. A convolutional neural network (CNN) end-to-end classifier was utilized to classify OCTA images from healthy control subjects and diabetic patients with no retinopathy (NoDR) and non-proliferative diabetic retinopathy (NPDR). For each eye, three en-face OCTA images were acquired from the superficial capillary plexus (SCP), deep capillary plexus (DCP), and choriocapillaris (CC) layers. The performances of the CNN classifier with individual layer inputs and multi-layer fusion architectures, including early-fusion, intermediate-fusion, and late-fusion, were quantitatively compared. For individual layer inputs, the superficial OCTA was observed to have the best performance, with 87.25% accuracy, 78.26% sensitivity, and 90.10% specificity, to differentiate control, NoDR, and NPDR. For multi-layer fusion options, the best option is the intermediate-fusion architecture, which achieved 92.65% accuracy, 87.01% sensitivity, and 94.37% specificity. To interpret the deep learning performance, the Gradient-weighted Class Activation Mapping (Grad-CAM) was utilized to identify spatial characteristics for OCTA classification. Comparative analysis indicates that the layer data fusion options can affect the performance of deep learning classification, and the intermediate-fusion approach is optimal for OCTA classification of DR.

To evaluate the efficacy of optical coherence tomography (OCT) angiography versus fluorescein angiography (FA) in terms of retinal vessel imaging in ischaemic diabetic maculopathy defined according to the Early Treatment Diabetic Retinopathy Study (ETDRS) classification. Twenty patients (31 eyes) with ischaemic diabetic maculopathy and 17 control subjects (27 eyes) were enrolled in this prospective study. Patients and control subjects underwent complete ophthalmic examination, including best-corrected visual acuity (BCVA), intraocular pressure, FA, Fourier domain optical coherence tomography (FD-OCT) and OCT angiography. Fluorescein angiograms and OCT angiography images were graded according to the foveal avascular zone (FAZ) of the ETDRS group. Ganglion cell complex (GCC) thickness was evaluated with FD-OCT. Optical coherence tomography (OCT) angiography images closely correlated with FA in terms of FAZ parameters. The correlation was strongest with OCT angiography deep imaging. The average GCC thickness was smaller in patients than in controls. Neither GCC parameters nor FAZ was correlated to BCVA. Given the correlation between FA and OCT angiography in terms of FAZ parameters, the newer method can be considered a valid, reliable and easy-to-perform method with which to evaluate ischaemic diabetic maculopathy without contrast injection, and thus to visualize and quantify non-perfusion areas without risks of anaphylactic reactions.

PurposeThis study aimed to evaluate the frequency of SSPiM (suspended scattering particles in motion), systemic risk factors, ocular findings, progression characteristics, and treatment response in diabetic retinopathy (DR) patients.MethodsIn this prospective study, a total of 109 eyes of 109 patients with diabetic macular edema (DME) were included. Demographic characteristics and systemic data of the patients were recorded. In addition to a detailed ophthalmological examination, optical coherence tomography (OCT) and OCT angiography (OCTA) imaging were performed. According to the OCTA images, the patients were divided into two categories: SSPiM detected (SSPiM +) and undetected (SSPiM −). The patients were followed up at 0, 3, and 6 months. Treatment responses at 6 months in treatment-administered patients with and without SSPiM were examined.ResultsThe frequency of SSPiM in DME cases was found to be 34.9%. No significant correlation was found between SSPiM and demographic characteristics, systemic, and biochemical parameters (p > 0.05). It was observed that SSPIM was most frequently localized in the outer nuclear layer adjacent to the outer plexiform (81.6%). SSPiM appearance disappeared in 7 (19.4%) of 36 patients with SSPiM who had regular follow-up for 6 months. In 4 (11.1%) of these seven patients, hard exudate plaques developed in the areas where SSPiM disappeared. Regarding treatment response at 6 months, the decrease in CMT was statistically significantly lower in the SSPiM group compared to cases without SSPiM.ConclusionSSPiM is a finding seen in approximately one-third of DME patients and may adversely affect the response to the treatment.

This paper introduces a new computer-aided diagnosis (CAD) system for detecting early-stage diabetic retinopathy (DR) using optical coherence tomography angiography (OCTA) images. The proposed DR-CAD system is based on the analysis of new local features that describe both the appearance and retinal structure in OCTA images. It starts with a new segmentation approach that has the ability to extract the blood vessels from superficial and deep retinal OCTA maps. The high capability of our segmentation approach stems from using a joint Markov-Gibbs random field stochastic model integrating a 3D spatial statistical model with a first-order appearance model of the blood vessels. Following the segmentation step, three new local features are estimated from the segmented vessels and the foveal avascular zone (FAZ): (a) vessels density, (b) blood vessel calibre, and (c) width of the FAZ. To distinguish mild DR patients from normal cases, the estimated three features are used to train and test a support vector machine (SVM) classifier with the radial basis function (RBF) kernel. On a cohort of 105 subjects, the presented DR-CAD system demonstrated an overall accuracy (ACC) of 94.3%, a sensitivity of 97.9%, a specificity of 87.0%, the area under the curve (AUC) of 92.4%, and a Dice similarity coefficient (DSC) of 95.8%. This in turn demonstrates the promise of the proposed CAD system as a supplemental tool for early detection of DR. We developed a new DR-CAD system that is capable of diagnosing DR in its early stage. The proposed system is based on extracting three different features from the segmented OCTA images, which reflect the changes in the retinal vasculature network.

High quality visualization of the retinal microvasculature can improve our understanding of the onset and development of retinal vascular diseases, especially Diabetic Retinopathy (DR), which is a major cause of visual morbidity and is increasing in prevalence. Optical Coherence Tomography Angiography (OCT-A) images are acquired over multiple seconds and are particularly susceptible to motion artifacts, which are more prevalent when imaging individuals with DR whose ability to fixate is limited due to deteriorating vision. The sequential acquisition and averaging of multiple OCT-A images can be performed for removing motion artifact and increasing the contrast of the vascular network. As motion artifacts often irreversibly corrupt OCT-A images of DR eyes, a robust registration pipeline is needed before feature preserving image averaging can be performed. In this report we present an improvement upon a novel method for the acquisition, processing, segmentation, registration, and averaging of sequentially acquired OCT-A images, to correct for motion artifacts in images of DR eyes. Image discontinuities caused by rapid micro-saccadic movements and image warping due to smoother reflex movements were corrected by strip-wise affine registration and subsequent local similarity-based non-rigid registration. Where our previous work was limited by the need for at least one image containing no motion artifact, thus reducing its clinical relevance, this novel template-less method stitches together partial images to form complete, motion-free images. These techniques significantly improve image quality, increasing the value for clinical diagnosis and increasing the range of patients for whom high quality OCT-A images can be acquired.

Objective As an extension of optical coherence tomography (OCT), optical coherence tomographic angiography (OCTA) provides information on the blood flow status at the microlevel and is sensitive to changes in the fundus vessels. However, due to the distinct imaging mechanism of OCTA, existing models, which are primarily used for analyzing fundus images, do not work well on OCTA images. Effectively extracting and analyzing the information in OCTA images remains challenging. To this end, a deep learning framework that fuses multilevel information in OCTA images is proposed in this study. The effectiveness of the proposed model was demonstrated in the task of diabetic retinopathy (DR) classification. Method First, a U-Net-based segmentation model was proposed to label the boundaries of large retinal vessels and the foveal avascular zone (FAZ) in OCTA images. Then, we designed an isolated concatenated block (ICB) structure to extract and fuse information from the original OCTA images and segmentation results at different fusion levels. Results The experiments were conducted on 301 OCTA images. Of these images, 244 were labeled by ophthalmologists as normal images, and 57 were labeled as DR images. An accuracy of 93.1% and a mean intersection over union (mIOU) of 77.1% were achieved using the proposed large vessel and FAZ segmentation model. In the ablation experiment with 6-fold validation, the proposed deep learning framework that combines the proposed isolated and concatenated convolution process significantly improved the DR diagnosis accuracy. Moreover, inputting the merged images of the original OCTA images and segmentation results further improved the model performance. Finally, a DR diagnosis accuracy of 88.1% (95%CI ± 3.6%) and an area under the curve (AUC) of 0.92 were achieved using our proposed classification model, which significantly outperforms the state-of-the-art classification models. As a comparison, an accuracy of 83.7 (95%CI ± 1.5%) and AUC of 0.76 were obtained using EfficientNet. Significance. The visualization results show that the FAZ and the vascular region close to the FAZ provide more information for the model than the farther surrounding area. Furthermore, this study demonstrates that a clinically sophisticated designed deep learning model is not only able to effectively assist in the diagnosis but also help to locate new indicators for certain illnesses.

Objective. Retinal vessel segmentation from optical coherence tomography angiography (OCTA) volumes is significant in analyzing blood supply structures and the diagnosing ophthalmic diseases. However, accurate retinal vessel segmentation in 3D OCTA remains challenging due to the interference of choroidal blood flow signals and the variations in retinal vessel structure. Approach. This paper proposes a layer attention network (LA-Net) for 3D-to-2D retinal vessel segmentation. The network comprises a 3D projection path and a 2D segmentation path. The key component in the 3D path is the proposed multi-scale layer attention module, which effectively learns the layer features of OCT and OCTA to attend to the retinal vessel layer while suppressing the choroidal vessel layer. This module also efficiently captures 3D multi-scale information for improved semantic understanding during projection. In the 2D path, a reverse boundary attention module is introduced to explore and preserve boundary and shape features of retinal vessels by focusing on non-salient regions in deep features. Main results. Experimental results in two subsets of the OCTA-500 dataset showed that our method achieves advanced segmentation performance with Dice similarity coefficients of 93.04% and 89.74%, respectively. Significance. The proposed network provides reliable 3D-to-2D segmentation of retinal vessels, with potential for application in various segmentation tasks that involve projecting the input image. Implementation code: https://github.com/y8421036/LA-Net.

PurposeTo develop an automated diabetic retinopathy (DR) staging system using optical coherence tomography angiography (OCTA) images with a convolutional neural network (CNN) and to verify the feasibility of the system.MethodsIn this retrospective cross-sectional study, a total of 918 data sets of 3 × 3 mm2 OCTA images and 917 data sets of 6 × 6 mm2 OCTA images were obtained from 1118 eyes. A deep CNN and four traditional machine learning models were trained with annotations made by a retinal specialist based on ultra-widefield fluorescein angiography. Separately, the same images of the test data sets were independently graded by two human experts. The results of the CNN algorithm were compared with those of traditional machine learning–based classifiers and human experts.ResultsThe proposed CNN achieved an accuracy of 0.728, a sensitivity of 0.675, a specificity of 0.944, an F1 score of 0.683, and a quadratic weighted κ of 0.908 for a six-level staging task, which were far superior to the results of traditional machine learning methods or human experts. The CNN algorithm showed a better performance using 6 × 6 mm2 rather than 3 × 3 mm2 sized OCTA images and using combined data rather than a separate OCTA layer alone.ConclusionsCNN-based classification using OCTA images can provide reliable assistance to clinicians for DR classification.Translational RelevanceThis CNN algorithm can guide the clinical decision for invasive angiography or referrals to ophthalmology specialists, helping to create more efficient diagnostic workflow in primary care settings.

Objective: To analyze optical coherence tomography angiography (OCTA) imaging of type 3 neovascularization in age-related macular degeneration (AMD) at baseline and following serial anti-vascular endothelial growth factor (anti-VEGF) treatments. Methods: This retrospective case series describes three treatment-naive patients diagnosed with type 3 neovascularization secondary to AMD based on clinical examination, fluorescein angiography (FA), and spectraldomain optical coherence tomography (SD-OCT). Written informed consent was obtained from all participants and approved by the Institutional Review Board of Northwestern University. Visual acuity and OCTA imaging with quantitative analysis of the type 3 neovascular complex was obtained at baseline and following monthly intravitreal anti-VEGF injections. Results: OCTA demonstrated resolution of cystoid macular edema in all three cases following anti-VEGF treatment. In one patient, resolution of the edema allowed enhanced visualization of the type 3 neovascular lesion due to intraretinal fluid obscuration at baseline. One case demonstrated persistence of larger vessels even after multiple anti-VEGF treatments. All cases showed improvement in visual acuity and reduction of type 3 neovascularization area on quantitative OCTA analysis. Conclusion: OCTA analysis of type 3 neovascularization demonstrated regression of small caliber vessels following longitudinal anti-VEGF treatment. Cystoid macular edema resolved and visual acuity improved in all cases. OCTA supplements fluorescein angiography and spectral domain OCT by providing improved microvascular identification of type 3 lesions and treatment response which may help guide clinician management and patient expectations.

Artifacts can affect optical coherence tomographic angiography (OCTA) images and may be associated with misinterpretation of OCT scans in both clinical trials and clinical settings. To identify the prevalence and type of artifacts in OCTA images associated with quantitative output and to analyze the role of proprietary quality indices in establishing image reliability. This cross-sectional study evaluated baseline OCTA images acquired in multicenter clinical trials and submitted to the Fundus Photograph Reading Center in Madison, Wisconsin, between January 1, 2016, and December 31, 2018. Images were captured using the 3 mm × 3 mm and/or 6 mm × 6 mm scan protocol with commercially available OCTA systems. Artifacts, including decentration, segmentation error, movement, blink, refraction shift, defocus, shadow, Z offset, tilt, and projection, were given a severity grade based on involvement of cross-sectional OCT and area of OCT grid affected. Prevalence and severity of OCTA artifacts and area under the receiver operating characteristic curve (AUC) of quality indices with image reliability. A total of 406 OCTA images from 234 eyes were included in this study, of which 221 (54.4%) were 6 mm × 6 mm scans and 185 (45.6%) were 3 mm × 3 mm scans. At least 1 artifact was documented in 395 images (97.3%). Severe artifacts associated with the reliability of quantitative outputs were found in 217 images (53.5%). Shadow (26.9% [109 images]), defocus (20.9% [85 images]), and movement (16.0% [65 images]) were the 3 most prevalent artifacts. Prevalence of artifacts did not vary with the imaging system used or with the scan protocol; however, the type of artifacts varied. Commercially recommended quality index thresholds had an AUC of 0.80 to 0.83, sensitivity of 97% to 99%, and specificity of 37% to 41% for reliable images. Findings from this study suggest that artifacts associated with quantitative outputs on commercially available OCTA devices are highly prevalent and that identifying common artifacts may require correlation with the angiogram and cross-sectional OCT scans. Knowledge of these artifacts and their implications for OCTA indices appears to be warranted for more accurate interpretation of OCTA images.