Fully Automated Detection and Quantification of Macular Fluid in OCT Using Deep Learning.

PURPOSE: Development and validation of a fully automated method to detect and quantify macular fluid in conventional OCT images. DESIGN: Development of a diagnostic modality. PARTICIPANTS: The clinical dataset for fluid detection consisted of 1200 OCT volumes of patients with neovascular age-related macular degeneration (AMD, n = 400), diabetic macular edema (DME, n = 400), or retinal vein occlusion (RVO, n = 400) acquired with Zeiss Cirrus (Carl Zeiss Meditec, Dublin, CA) (n = 600) or Heidelberg Spectralis (Heidelberg Engineering, Heidelberg, Germany) (n = 600) OCT devices. METHODS: A method based on deep learning to automatically detect and quantify intraretinal cystoid fluid (IRC) and subretinal fluid (SRF) was developed. The performance of the algorithm in accurately identifying fluid localization and extent was evaluated against a manual consensus reading of 2 masked reading center graders. MAIN OUTCOME MEASURES: Performance of a fully automated method to accurately detect, differentiate, and quantify intraretinal and SRF using area under the receiver operating characteristics curves, precision, and recall. RESULTS: The newly designed, fully automated diagnostic method based on deep learning achieved optimal accuracy for the detection and quantification of IRC for all 3 macular pathologies with a mean accuracy (AUC) of 0.94 (range, 0.91-0.97), a mean precision of 0.91, and a mean recall of 0.84. The detection and measurement of SRF were also highly accurate with an AUC of 0.92 (range, 0.86-0.98), a mean precision of 0.61, and a mean recall of 0.81, with superior performance in neovascular AMD and RVO compared with DME, which was represented rarely in the population studied. High linear correlation was confirmed between automated and manual fluid localization and quantification, yielding an average Pearson's correlation coefficient of 0.90 for IRC and of 0.96 for SRF. CONCLUSIONS: Deep learning in retinal image analysis achieves excellent accuracy for the differential detection of retinal fluid types across the most prevalent exudative macular diseases and OCT devices. Furthermore, quantification of fluid achieves a high level of concordance with manual expert assessment. Fully automated analysis of retinal OCT images from clinical routine provides a promising horizon in improving accuracy and reliability of retinal diagnosis for research and clinical practice in ophthalmology.

Neuroretinal atrophy following resolution of macular oedema in retinal vein occlusion.

BACKGROUND/AIMS: To characterise neuroretinal atrophy in retinal vein occlusion (RVO). METHODS: We included patients with central/branch RVO (CRVO=196, BRVO=107) who received ranibizumab according to a standardised protocol for 6 months. Retinal atrophy was defined as the presence of an area of retinal thickness (RT) <260 µm outside the foveal centre. Moreover, the thickness of three distinct retinal layer compartments was computed as follows: (1) retinal nerve fibre layer to ganglion cell layer, (2) inner plexiform layer (IPL) to outer nuclear layer (ONL) and (3) inner segment/outer segment junction to retinal pigment epithelium. To characterise atrophy further, we assessed perfusion status on fluorescein angiography and best-corrected visual acuity (BCVA), and compared these between eyes with/without atrophy. RESULTS: 23 patients with CRVO and 11 patients with BRVO demonstrated retinal atrophy, presenting as sharply demarcated retinal thinning confined to a macular quadrant. The mean RT in the atrophic quadrant at month 6 was 249±26 µm (CRVO) and 244±29 µm (BRVO). Individual layer analysis revealed pronounced thinning in the IPL to ONL compartment. Change in BCVA at 6 months was similar between the groups (BRVO, +15 vs +18 letters; CRVO, +14 vs +18 letters). CONCLUSIONS: In this exploratory analysis, we describe the characteristics of neuroretinal atrophy in RVO eyes with resolved macular oedema after ranibizumab therapy. Our analysis shows significant, predominantly retinal thinning in the IPL to ONL compartment in focal macular areas in 11% of patients with RVO. Eyes with retinal atrophy did not show poorer BCVA outcomes.

Evaluating the impact of vitreomacular adhesion on anti-VEGF therapy for retinal vein occlusion using machine learning.

Vitreomacular adhesion (VMA) represents a prognostic biomarker in the management of exudative macular disease using anti-vascular endothelial growth factor (VEGF) agents. However, manual evaluation of VMA in 3D optical coherence tomography (OCT) is laborious and data on its impact on therapy of retinal vein occlusion (RVO) are limited. The aim of this study was to (1) develop a fully automated segmentation algorithm for the posterior vitreous boundary and (2) to study the effect of VMA on anti-VEGF therapy for RVO. A combined machine learning/graph cut segmentation algorithm for the posterior vitreous boundary was designed and evaluated. 391 patients with central/branch RVO under standardized ranibizumab treatment for 6/12 months were included in a systematic post-hoc analysis. VMA (70%) was automatically differentiated from non-VMA (30%) using the developed method combined with unsupervised clustering. In this proof-of-principle study, eyes with VMA showed larger BCVA gains than non-VMA eyes (BRVO: 15 ± 12 vs. 11 ± 11 letters, p = 0.02; CRVO: 18 ± 14 vs. 9 ± 13 letters, p < 0.01) and received a similar number of retreatments. However, this association diminished after adjustment for baseline BCVA, also when using more fine-grained VMA classes. Our study illustrates that machine learning represents a promising path to assess imaging biomarkers in OCT.

Impact of B-Scan Averaging on Spectralis Optical Coherence Tomography Image Quality before and after Cataract Surgery.

BACKGROUND AND OBJECTIVE: To determine optimal image averaging settings for Spectralis optical coherence tomography (OCT) in patients with and without cataract. STUDY DESIGN/MATERIAL AND METHODS: In a prospective study, the eyes were imaged before and after cataract surgery using seven different image averaging settings. Image quality was quantitatively evaluated using signal-to-noise ratio, distinction between retinal layer image intensity distributions, and retinal layer segmentation performance. Measures were compared pre- and postoperatively across different degrees of averaging. RESULTS: 13 eyes of 13 patients were included and 1092 layer boundaries analyzed. Preoperatively, increasing image averaging led to a logarithmic growth in all image quality measures up to 96 frames. Postoperatively, increasing averaging beyond 16 images resulted in a plateau without further benefits to image quality. Averaging 16 frames postoperatively provided comparable image quality to 96 frames preoperatively. CONCLUSION: In patients with clear media, averaging 16 images provided optimal signal quality. A further increase in averaging was only beneficial in the eyes with senile cataract. However, prolonged acquisition time and possible loss of details have to be taken into account.

Spatial Correspondence Between Intraretinal Fluid, Subretinal Fluid, and Pigment Epithelial Detachment in Neovascular Age-Related Macular Degeneration.

Purpose: To identify the spatial distribution of exudative features of choroidal neovascularization in neovascular age-related macular degeneration (nAMD) based on the localization of intraretinal cystoid fluid (IRC), subretinal fluid (SRF), and pigment-epithelial detachment (PED). Methods: This retrospective cross-sectional study included spectral-domain optical coherence tomography volume scans (6 × 6 mm) of 1341 patients with treatment-naïve nAMD. IRC, SRF, and PED were detected on a per-voxel basis using fully automated segmentation algorithms. Two subsets of 37 volumes each were manually segmented to validate the automated results. The spatial correspondence of components was quantified by computing proportions of IRC-, SRF-, or PED-presenting A-scans simultaneously affected by the respective other pathomorphologic components on a per-patient basis. The median across the population is reported. Odds ratios between pairs of lesions were calculated and tested for significance pixel wise. Results: Automated image segmentation was successful in 1182 optical coherence tomography volumes, yielding more than 61 million A-scans for analysis. Overall, 81% of eyes showed IRC, 95% showed SRF, and 92% showed PED. IRC-presenting A-scans also showed SRF in a median 2.5%, PED in 32.9%. Of the SRF-presenting A-scans, 0.3% demonstrated IRC, 1.4% PED. Of the PED-presenting A-scans, 5.2% contained IRC, 2.0% SRF. Similar patterns were observed in the manually segmented subsets and via pixel-wise odds ratio analysis. Conclusions: Automated analyses of large-scale datasets in a cross-sectional study of 1182 patients with active treatment-naïve nAMD demonstrated low spatial correlation of SRF with IRC and PED in contrast to increased colocalization of IRC and PED. These morphological associations may contribute to our understanding of functional deficits in nAMD.

Correlation of 3-Dimensionally Quantified Intraretinal and Subretinal Fluid With Visual Acuity in Neovascular Age-Related Macular Degeneration.

IMPORTANCE: Robust and sensitive imaging biomarkers for visual function are an unmet medical need in the management of neovascular age-related macular degeneration. OBJECTIVE: To determine the correlation of 3-dimensionally quantified intraretinal cystoid fluid (IRC) and subretinal fluid (SRF) with best-corrected visual acuity (BCVA) in treatment-naive neovascular age-related macular degeneration and during antiangiogenic therapy. DESIGN, SETTING, AND PARTICIPANTS: Retrospective cohort study between November 2009 and November 2011 at an institutional referral center and reading center of patients with treatment-naive subfoveal choroidal neovascularization receiving intravitreal ranibizumab or aflibercept over 12 months. All individual IRC and SRF lesions were manually delineated on each of the 128 B-scan sections of spectral-domain optical coherence tomographic volume scans at baseline and months 1, 6, and 12. Correlations were computed between the IRC and SRF parameters and the baseline BCVA, final BCVA, and BCVA change. A systematic parameter search was conducted to detect annotation-derived variables with best predictive value. An exponential model for BCVA change balancing for the ceiling effect was constructed. MAIN OUTCOMES AND MEASURES: Goodness of fit of correlations between the IRC and SRF parameters and the baseline BCVA, final BCVA, and BCVA change. RESULTS: Thirty-eight patients were included (25 female, 13 male; mean [SD] age at enrollment, 78.49 [8.23] years; mean [SD] BCVA score at baseline, 54 [16] Early Treatment Diabetic Retinopathy Study letters [Snellen equivalent approximately 20/160], with a gain to 63 [19] letters [Snellen equivalent approximately 20/100] at month 12). A total of 19,456 scans underwent complete quantification of IRC and SRF. The best correlation with BCVA at baseline was achieved using a coverage-based, foveal area-weighted IRC parameter (R2 = 0.59; P < .001). The same baseline parameter also predicted BCVA at 12 months (R2 = 0.21; P = .003). The BCVA gain correlated with IRC decrease in the exponential model (R2 = 0.40; P < .001) and linear model (R2 = 0.25; P = .002). No robust associations were found between SRF and baseline BCVA (R2 = 0.06; P = .14) or BCVA change (R2 = 0.14; P = .02). CONCLUSIONS AND RELEVANCE: In this proof-of-principle study, IRC-derived morphometric variables correlated well with treatment-naive BCVA and BCVA outcomes in antiangiogenic therapy. While IRC reduction was associated with BCVA gains, some IRC-mediated neurosensory damage remained permanent.

A novel benchmark model for intelligent annotation of spectral-domain optical coherence tomography scans using the example of cyst annotation.

BACKGROUND AND OBJECTIVES: The lack of benchmark data in computational ophthalmology contributes to the challenging task of applying disease assessment and evaluate performance of machine learning based methods on retinal spectral domain optical coherence tomography (SD-OCT) scans. Presented here is a general framework for constructing a benchmark dataset for retinal image processing tasks such as cyst, vessel, and subretinal fluid segmentation and as a result, a benchmark dataset for cyst segmentation has been developed. METHOD: First, a dataset captured by different SD-OCT vendors with different numbers of scans and pathology qualities are selected. Then a robust and intelligent method is used to evaluate performance of readers, partitioning the dataset into subsets. Subsets are then assigned to complementary readers for annotation with respect to a novel confidence based annotation protocol. Finally, reader annotations are combined based on their performance to generate final annotations. RESULT: The generated benchmark dataset for cyst segmentation comprises 26 SD-OCT scans with differing cyst qualities, collected from 4 different SD-OCT vendors to cover a wide variety of data. The dataset is partitioned into three subsets which are annotated by complementary readers based on a confidence based annotation protocol. Experimental results show annotations of complementary readers are combined efficiently with respect to their performance, generating accurate annotations. CONCLUSION: Our results facilitate the process of generating benchmark datasets. Moreover the generated benchmark data set for cyst segmentation can be used reliably to train and test machine learning based methods.

Multivendor Spectral-Domain Optical Coherence Tomography Dataset, Observer Annotation Performance Evaluation, and Standardized Evaluation Framework for Intraretinal Cystoid Fluid Segmentation.

Development of image analysis and machine learning methods for segmentation of clinically significant pathology in retinal spectral-domain optical coherence tomography (SD-OCT), used in disease detection and prediction, is limited due to the availability of expertly annotated reference data. Retinal segmentation methods use datasets that either are not publicly available, come from only one device, or use different evaluation methodologies making them difficult to compare. Thus we present and evaluate a multiple expert annotated reference dataset for the problem of intraretinal cystoid fluid (IRF) segmentation, a key indicator in exudative macular disease. In addition, a standardized framework for segmentation accuracy evaluation, applicable to other pathological structures, is presented. Integral to this work is the dataset used which must be fit for purpose for IRF segmentation algorithm training and testing. We describe here a multivendor dataset comprised of 30 scans. Each OCT scan for system training has been annotated by multiple graders using a proprietary system. Evaluation of the intergrader annotations shows a good correlation, thus making the reproducibly annotated scans suitable for the training and validation of image processing and machine learning based segmentation methods. The dataset will be made publicly available in the form of a segmentation Grand Challenge.

Choroidal thickness maps from spectral domain and swept source optical coherence tomography: algorithmic versus ground truth annotation.

BACKGROUND/AIMS: The purpose of the study was to create a standardised protocol for choroidal thickness measurements and to determine whether choroidal thickness measurements made on images obtained by spectral domain optical coherence tomography (SD-OCT) and swept source (SS-) OCT from patients with healthy retina are interchangeable when performed manually or with an automatic algorithm. METHODS: 36 grid cell measurements for choroidal thickness for each volumetric scan were obtained, which were measured for SD-OCT and SS-OCT with two methods on 18 eyes of healthy volunteers. Manual segmentation by experienced retinal graders from the Vienna Reading Center and automated segmentation on >6300 images of the choroid from both devices were statistically compared. RESULTS: Model-based comparison between SD-OCT/SS-OCT showed a systematic difference in choroidal thickness of 16.26±0.725 µm (p<0.001) for manual segmentation and 21.55±0.725 µm (p<0.001) for automated segmentation. Comparison of automated with manual segmentations revealed small differences in thickness of -0.68±0.513 µm (p=0.1833). The correlation coefficients for SD-OCT and SS-OCT measures within eyes were 0.975 for manual segmentation and 0.955 for automatic segmentation. CONCLUSION: Choroidal thickness measurements of SD-OCT and SS-OCT indicate that these two devices are interchangeable with a trend of choroidal thickness measurements being slightly thicker on SD-OCT with limited clinical relevance. Use of an automated algorithm to segment choroidal thickness was validated in healthy volunteers.