Repeatability of binarization thresholding methods for optical coherence tomography angiography image quantification.

Binarization is a critical step in analysis of retinal optical coherence tomography angiography (OCTA) images, but the repeatability of metrics produced from various binarization methods has not been fully assessed. This study set out to examine the repeatability of OCTA quantification metrics produced using different binarization thresholding methods, all of which have been applied in previous studies, across multiple devices and plexuses. Successive 3 × 3 mm foveal OCTA images of 13 healthy eyes were obtained on three different devices. For each image, contrast adjustments, 3 image processing techniques (linear registration, histogram normalization, and contrast-limited adaptive histogram equalization), and 11 binarization thresholding methods were independently applied. Vessel area density (VAD) and vessel length were calculated for retinal vascular images. Choriocapillaris (CC) images were quantified for VAD and flow deficit metrics. Repeatability, measured using the intra-class correlation coefficient, was inconsistent and generally not high (ICC < 0.8) across binarization thresholds, devices, and plexuses. In retinal vascular images, local thresholds tended to incorrectly binarize the foveal avascular zone as white (i.e., wrongly indicating flow). No image processing technique analyzed consistently resulted in highly repeatable metrics. Across contrast changes, retinal vascular images showed the lowest repeatability and CC images showed the highest.

Using the Pathophysiology of Dry AMD to Guide Binarization of the Choriocapillaris on OCTA: A Model.

Especially since the incorporation of swept laser sources, optical coherence tomography angiography (OCTA) has enabled quantification of choriocapillaris perfusion. A critical step in this process is binarization, which makes angiographic images quantifiable in terms of perfusion metrics. It remains challenging to have confidence that choriocapillaris perfusion metrics reflect the reality of pathophysiologic flow, largely because choice of binarization method can result in significantly different perfusion metric outcomes. This commentary discusses a proof-of-concept case involving comparative assessment of binarization methods for a set of dry age-related macular degeneration OCTA data. One of these methods was deemed preferable based on superior agreement with suspected physiologic and pathophysiologic characteristics, thus demonstrating the principle that, in the absence of gold standards for measurement of choriocapillaris perfusion, the best available approximations of pathophysiology may be used to guide choice of binarization method.

Model-to-Data Approach for Deep Learning in Optical Coherence Tomography Intraretinal Fluid Segmentation.

Importance: Amid an explosion of interest in deep learning in medicine, including within ophthalmology, concerns regarding data privacy, security, and sharing are of increasing importance. A model-to-data approach, in which the model itself is transferred rather than data, can circumvent many of these challenges but has not been previously demonstrated in ophthalmology. Objective: To determine whether a model-to-data deep learning approach (ie, validation of the algorithm without any data transfer) can be applied in ophthalmology. Design, Setting, and Participants: This single-center cross-sectional study included patients with active exudative age-related macular degeneration undergoing optical coherence tomography (OCT) at the New England Eye Center from August 1, 2018, to February 28, 2019. Data were primarily analyzed from March 1 to June 20, 2019. Main Outcomes and Measures: Training of the deep learning model, using a model-to-data approach, in recognizing intraretinal fluid (IRF) on OCT B-scans. Results: The model was trained (learning curve Dice coefficient, >80%) using 400 OCT B-scans from 128 participants (69 female [54%] and 59 male [46%]; mean [SD] age, 77.5 [9.1] years). In comparing the model with manual human grading of IRF pockets, no statistically significant difference in Dice coefficients or intersection over union scores was found (P > .05). Conclusions and Relevance: A model-to-data approach to deep learning applied in ophthalmology avoided many of the traditional hurdles in large-scale deep learning, including data sharing, security, and privacy concerns. Although the clinical relevance of these results is limited at this time, this proof-of-concept study suggests that such a paradigm should be further examined in larger-scale, multicenter deep learning studies.

Repeatability and Reproducibility of Photoreceptor Density Measurement in the Macula Using the Spectralis High Magnification Module.

PURPOSE: To evaluate the repeatability and reproducibility of photoreceptor density assessment with manual cell counting in healthy participants imaged with the Heidelberg Spectralis High Magnification Module (HMM). DESIGN: Precision study, evaluation of diagnostic test or technology. PARTICIPANTS: Eleven eyes of 8 participants. METHODS: Images were acquired using the Spectralis HMM by a single operator during 2 separate imaging sessions. The 3 highest-quality images of each eye from each session were selected for analysis and coregistered. For a subset of participants, a second operator acquired images in 1 session, and images with the best quality were selected for analysis. Photoreceptor densities were obtained by manual counts in squares of 0.0625 mm2 located in the parafovea. Repeatability (intragrader and intrasession) and reproducibility (interoperator, intergrader, and intersession) were assessed by calculating the intraclass correlation coefficient (ICC) from linear mixed effects models. Bland-Altman plots, coefficients of repeatability, and Pearson correlation results were reported. MAIN OUTCOME MEASURES: Intragrader, intrasession, intersession, interoperator, and intergrader ICC estimates and their 95% confidence intervals for photoreceptor density measurements in the parafovea. RESULTS: Twenty-four eyes of 13 healthy participants were imaged initially. Of these, 11 eyes (45.83%) of 8 participants that had at least 3 acceptable images in each session were included in this study. Mean parafoveal photoreceptor density was 14 988 cells/mm2 (standard deviation, 1403.15 cells/mm2). Intragrader ICC was 0.84 (95% confidence interval, 0.57-0.95), intrasession ICC was 0.69 (95% confidence interval, 0.17-0.86), intersession ICC was 0.88 (95% confidence interval, 0.53-0.96), interoperator ICC was 0.70 (95% confidence interval, 0-0.95), and intergrader ICC was 0.22 (95% confidence interval, 0-0.71). CONCLUSIONS: Images obtained with the HMM allow for photoreceptor mosaic visualization in the macular area, mainly in the parafovea. Although densities obtained are in accordance with other reported methods in the literature, variability within and between images of the apparent cell mosaic were observed, and this study did not demonstrate high repeatability or reproducibility for quantitative assessments using the manual counting method.

Topographic analysis of macular choriocapillaris flow deficits in diabetic retinopathy using swept-source optical coherence tomography angiography.

BACKGROUND: The purpose of this study was to investigate the association between diabetic retinopathy (DR) severity and macular choriocapillaris (CC) flow deficit percentage (FD %) in different macular regions using swept-source optical coherence tomography angiography (SS-OCTA). METHODS: Diabetic patients with SS-OCTA images were graded by severity and retrospectively assessed. CC FD % was calculated in four different regions of the OCTA image: inner, middle, outer, and full-field region. The generalized estimating equations (GEE) approach for clustered eye data was used to determine effect size and significance of age and disease severity on FD % for each region. RESULTS: 160 eyes from 90 total diabetic patients met inclusion criteria. Out of 90 patients, 33 had no DR, 17 had mild nonproliferative DR (NPDR), 8 had moderate NPDR, 10 had severe NPDR and 22 had proliferative DR. Age and DR severity had a significant positive association with FD % for each region studied with a greater effect in the two centermost regions. The increase in flow deficit percentage per year of age by region was: inner 0.12 (p < 0.001), middle 0.09 (p < 0.001), outer 0.05 (p < 0.001, full-field 0.06 (p < 0.001). The increase in flow deficit percentage per increase in diabetic retinopathy severity stage by region was: inner 0.65 (p < 0.0087), middle 0.56 (p < 0.0012), outer 0.33 (p < 0.045), full-field 0.36 (p < 0.018). CONCLUSIONS: Topographic analysis of the CC FD % in diabetic eyes suggests that CC flow impairment corresponds to DR severity, with all studied regions of the CC significantly affected. There was greater regional impairment due to age and disease severity in the inner and middle regions.

Global Analysis of Macular Choriocapillaris Perfusion in Dry Age-Related Macular Degeneration using Swept-Source Optical Coherence Tomography Angiography.

Purpose: Swept-source optical coherence tomography angiography (SS-OCTA) was used to investigate if the clinical stage of dry age-related macular degeneration (AMD) was correlated with global and regional macular choriocapillaris (CC) perfusion. Methods: In this retrospective, cross-sectional study, 6 × 6-mm SS-OCTA images from eyes with early, intermediate, and advanced dry AMD (56 eyes, 41 patients) were analyzed using algorithms described in the literature to assess regional flow deficit percentage (FD%) and average flow deficit size. Regions were defined by concentric areas centered on the fovea: a 1-mm-diameter area, 3-mm-diameter ring, 5-mm-diameter area, 5-mm-diameter ring, and 6 × 6-mm whole image. Data were modeled using the generalized estimating equations approach. Results: The relationship between age and CC FD% and average flow deficit size was statistically significant (P ≤ 0.05) in all regions of analysis by linear modeling. The relationship between dry AMD stage and FD% was statistically significant by linear modeling in the 5-mm ring, and between dry AMD stage and average flow deficit size in the 3-mm ring, 5-mm area, 5-mm ring, and 6 × 6-mm whole image. Conclusions: Linear modeling suggests a statistically significant relationship between dry AMD stage and CC perfusion, most prominent in the more peripheral regions of the macula.

Impact of Binarization Thresholding and Brightness/Contrast Adjustment Methodology on Optical Coherence Tomography Angiography Image Quantification.

PURPOSE: Binarization is a critical technique in optical coherence tomography angiography (OCTA) image analysis, but there is no consistency in the method used in published OCTA studies. This study assessed whether differences in OCTA binarization and brightness and contrast adjustments affect quantification measurements. DESIGN: Prospective cross-sectional validity study. METHODS: This was a single-center study examining 21 eyes of 11 healthy individuals. All eyes were imaged using a swept-source OCTA (Zeiss), and quantitative measurements resulting from five binarization thresholding and five brightness/contrast adjustment methods were compared. All measurements were calculated for the superficial plexus and choriocapillaris (CC), as well as unaveraged and averaged en face OCTA images. RESULTS: There were statistically significant differences between measurements from different binarization thresholding methods (P < 0.0001), as well as measurements from different histogram adjustments (P < 0.0001). The binarization thresholds yielded different measurements when combined with variable brightness/contrast adjustments. The method of analysis also affected the directionality of trends in imaging measurements between unaveraged and averaged CC images. CONCLUSIONS: The method of OCTA image binarization thresholding and histogram adjustment significantly alters quantitative measurements and the directionality of trends. Results obtained from different OCTA binarization methods should be seen as valid only for that given method. This has significant consequences for clinical trials using OCTA measurements as outcome measurements. A consensus is needed across the research community for a consistent method for OCTA image quantification and greater attention paid to fully describing methods in published studies.