Dataset Information

Automated detection of a nonperfusion area caused by retinal vein occlusion in optical coherence tomography angiography images using deep learning.

ABSTRACT: We aimed to assess the ability of deep learning (DL) and support vector machine (SVM) to detect a nonperfusion area (NPA) caused by retinal vein occlusion (RVO) with optical coherence tomography angiography (OCTA) images. The study included 322 OCTA images (normal: 148; NPA owing to RVO: 174 [128 branch RVO images and 46 central RVO images]). Training to construct the DL model using deep convolutional neural network (DNN) algorithms was provided using OCTA images. The SVM used a scikit-learn library with a radial basis function kernel. The area under the curve (AUC), sensitivity and specificity for detecting an NPA were examined. We compared the diagnostic ability (sensitivity, specificity and average required time) between the DNN, SVM and seven ophthalmologists. Heat maps were generated. With regard to the DNN, the mean AUC, sensitivity, specificity and average required time for distinguishing RVO OCTA images with an NPA from normal OCTA images were 0.986, 93.7%, 97.3% and 176.9 s, respectively. With regard to SVM, the mean AUC, sensitivity, and specificity were 0.880, 79.3%, and 81.1%, respectively. With regard to the seven ophthalmologists, the mean AUC, sensitivity, specificity and average required time were 0.962, 90.8%, 89.2%, and 700.6 s, respectively. The DNN focused on the foveal avascular zone and NPA in heat maps. The performance of the DNN was significantly better than that of SVM in all parameters (p < 0.01, all) and that of the ophthalmologists in AUC and specificity (p < 0.01, all). The combination of DL and OCTA images had high accuracy for the detection of an NPA, and it might be useful in clinical practice and retinal screening.

SUBMITTER: Nagasato D

PROVIDER: S-EPMC6837754 | biostudies-literature | 2019

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Automated detection of a nonperfusion area caused by retinal vein occlusion in optical coherence tomography angiography images using deep learning.

Nagasato Daisuke D Tabuchi Hitoshi H Masumoto Hiroki H Enno Hiroki H Ishitobi Naofumi N Kameoka Masahiro M Niki Masanori M Mitamura Yoshinori Y

PloS one 20191107 11

We aimed to assess the ability of deep learning (DL) and support vector machine (SVM) to detect a nonperfusion area (NPA) caused by retinal vein occlusion (RVO) with optical coherence tomography angiography (OCTA) images. The study included 322 OCTA images (normal: 148; NPA owing to RVO: 174 [128 branch RVO images and 46 central RVO images]). Training to construct the DL model using deep convolutional neural network (DNN) algorithms was provided using OCTA images. The SVM used a scikit-learn lib ...[more]

PMID: 31697697

Similar Datasets

Project description:ImportanceMacular ischemia is a key feature of diabetic retinopathy (DR). Quantification of macular ischemia has potential as a biomarker for DR.ObjectiveTo assess the feasibility of automated quantification of capillary nonperfusion as a potential sign of macular ischemia using optical coherence tomography (OCT) angiography.Design, setting, and participantsAn observational study conducted in a tertiary, subspecialty, academic practice evaluated macular nonperfusion with 6 × 6-mm OCT angiography obtained with commercially available 70-kHz OCT and fluorescein angiography (FA). The study was conducted from January 22 to September 18, 2014. Data analysis was performed from October 1, 2014, to April 7, 2015. Participants included 12 individuals with normal vision serving as controls and 12 patients with various levels of DR.Main outcomes and measuresPreplanned primary measures were parafoveal and perifoveal vessel density, total avascular area, and foveal avascular zone as detected with 6 × 6-mm OCT angiography and analyzed using an automated algorithm. Secondary measures included the agreement of the avascular area between the OCT angiogram and FA.ResultsCompared with the 12 healthy controls (11 women; mean [SD] age, 54.2 [14.2] years), the 12 participants with DR (4 women; mean [SD] age, 55.1 [12.1] years) had reduced parafoveal and perifoveal vessel density by 12.6% (95% CI, 7.7%-17.5%; P < .001) and 10.4% (95% CI, 6.8%-14.1%; P < .001), respectively. Total avascular area and foveal avascular zone area were greater in eyes with DR by 0.82 mm2 (95% CI, 0.65-0.99 mm2; P = .02) and 0.16 mm2 (95% CI, 0.05-0.28 mm2; P < .001). The agreement between the vascular areas in the OCT angiogram and FA had a κ value of 0.45 (95% CI, 0.21-0.70; P < .001). Total avascular area in the central 5.5-mm-diameter area distinguished eyes with DR from control eyes with 100% sensitivity and specificity.Conclusions and relevanceAvascular area analysis with an automated algorithm using OCT angiography, although not equivalent to FA, detected DR reliably in this small pilot study. Further study is necessary to determine the usefulness of the automated quantification in clinical practice.

Project description:PurposeTo examine whether deep-learning denoised optical coherence tomography angiography (OCTA) images could enhance automated macular ischemia quantification in branch retinal vein occlusion (BRVO).MethodsThis retrospective, single-center, cross-sectional study enrolled 74 patients with BRVO and 46 age-matched healthy subjects. The severity of macular ischemia was graded as mild, moderate, or severe. Denoised OCTA images were produced using a neural network model. Quantitative parameters derived from denoised images, including vessel density and nonperfusion area, were compared with those derived from the OCTA machine. The main outcome measures were correlations between quantitative parameters, and areas under receiver operating characteristic curves (AUCs) in classifying the severity of the macular ischemia.ResultsThe vessel density and nonperfusion area from denoised images were correlated strongly with the corresponding parameters from machine-derived images in control eyes and BRVO eyes with mild or moderate macular ischemia (all P < 0.001). However, no such correlation was found in eyes with severe macular ischemia. The vessel density and nonperfusion area from denoised images had significantly larger area under receiver operating characteristic curve than those derived from the original images in classifying moderate versus severe macular ischemia (0.927 vs 0.802 [P = 0.042] and 0.946 vs 0.797, [P = 0.022], respectively). There were no significant differences in the areas under receiver operating characteristic curve between the denoised images and the machine-derived parameters in classifying control versus BRVO, and mild versus moderate macular ischemia.ConclusionsA neural network model is useful for removing speckle noise on OCTA images and facilitating the automated grading of macular ischemia in eyes with BRVO.Translational relevanceDeep-learning denoised optical coherence tomography angiography images could enhance automated macular ischemia quantification.

Project description:BackgroundRetinal vein occlusion (RVO) is the second most common vision-threatening retinal vascular disease. Intravitreal dexamethasone implant has been applied to treat macular edema secondary to RVO (RVO-ME). However, the alteration of morphologic features detected with optical coherence tomography angiography (OCTA) has not been fully studied in RVO-ME patients before and after the treatment.ObjectiveThis study is aimed at identifying potential therapeutic targets in RVO with integrative bioinformatic analysis and compares the OCTA characteristics alterations in patients with RVO-ME receiving injection of dexamethasone intravitreal implant.MethodsBioinformatic analysis was analyzed in GSE101398 dataset from the Gene Expression Omnibus database. Multiple functional enrichment analyses were performed, and protein-protein interaction network was constructed to visualize the key node genes. Eleven eyes with RVO-ME were examined with OCTA before and after intravitreal dexamethasone implant. The OCTA parameters, including macular thickness, vessel density, foveal avascular zone parameters, the number of hyperreflective foci (HRF), area of cystoid edema, and subretinal fluid (SRF), were compared. The correlation was analyzed between best-corrected visual acuity (BCVA) and OCTA parameters.ResultsA total of 79 differentially expressed genes were identified. Functional enrichment analyses revealed the enriched inflammatory events in RVO. In RVO-ME, Pearson correlation revealed that baseline BCVA was positively correlated with the area of SRF and central macular thickness, while no correlation was detected between baseline BCVA and HRF number or the area of cystoid edema. The visual acuity improved, and the central macular thickness was decreased after intravitreal dexamethasone implant injection. Besides, the number of HRF, the area of cystoid edema, and SRF were significantly alleviated after dexamethasone intravitreal injection.ConclusionRetinal inflammation plays a crucial role in RVO pathogenesis. The imaging biomarkers of RVO including Müller glial intracellular edema, and retinal pigment epithelium dysfunction, could be assessed in OCTA and attenuated by intravitreal dexamethasone implant effectively.

Project description:Importance:Diabetic retinopathy (DR) is a leading cause of vision loss that is managed primarily through qualitative clinical examination of the retina. Optical coherence tomography angiography (OCTA) may offer an objective and quantitative method of evaluating DR. Objective:To quantify capillary nonperfusion in 3 vascular plexuses in the macula of eyes patients with diabetes of various retinopathy severity using projection-resolved OCTA (PR-OCTA). Design, Setting, and Participants:Cross-sectional study at a tertiary academic center comprising 1 eye each from healthy control individuals and patients with diabetes at different severity stages of retinopathy. Data were acquired and analyzed between January 2015 and December 2017. Main Outcomes and Measures:Foveal avascular zone area, extrafoveal avascular area (EAA), and the sensitivity of detecting levels of retinopathy. Results:The study included 39 control individuals (20 women [51%]; mean [SD] age, 43.41 [19.37] years); 16 patients with diabetes without retinopathy (8 women [50%]; mean [SD] age, 56.50 [12.43] years); 23 patients with mild to moderate nonproliferative DR (18 women [78%] ; mean [SD] age, 62.48 [10.55] years); and 32 patients with severe nonproliferative DR or proliferative DR (12 women [38%]; mean age, 53.41 [14.05] years). Mean (SD) foveal avascular zone area was 0.203 (0.103) mm2 for control individuals, 0.192 (0.084) mm2 for patients with diabetes without retinopathy, 0.243 [0.079] mm2 for mild to moderate nonproliferative DR, and 0.359 (0.275) mm2 for severe nonproliferative DR or proliferative DR. Mean (SD) EAA in whole inner retinal slab in these groups, respectively, were 0.020 (0.031) mm2, 0.034 (0.047) mm2, 0.038 (0.040) mm2, and 0.237 (0.235) mm2. The mean (SD) sum of EAA from 3 segmented plexuses in each of the respective groups were 0.103 (0.169) mm2, 0.213 (0.242) mm2, 0.451 (0.243) mm2, and 1.325 (1.140) mm2. With specificity fixed at 95%, using EAA in inner retinal slab, the sensitivity of detecting patients with diabetes from healthy control individuals was 28% (95% CI, 18%-40%), 31% for patients with DR (95% CI, 19%-45%), and 47% for patients with severe DR (95% CI, 29%-65%) from whole inner retinal EAA. With the sum of EAA from 3 individual plexuses, the sensitivities were 69% (95% CI, 57%-80%), 82% (95% CI, 70%-91%), and 97% (95% CI, 85%-100%), respectively. Avascular areas were not associated with signal strength index. The commercial vessel density from the 2-plexus scheme distinguished the groups with lower sensitivity and were dependent on SSI. Conclusions and Relevance:Automatically quantified avascular areas from a 3-layer segmentation scheme using PR-OCTA distinguished levels of retinopathy with a greater sensitivity than avascular areas from unsegmented inner retinal slab or measurements from a commercially available vessel density in 2-layer scheme. Additional studies are needed to investigate the applicability of nonperfusion parameters in clinical settings.

Dataset Information

Automated detection of a nonperfusion area caused by retinal vein occlusion in optical coherence tomography angiography images using deep learning.

Publications

Automated detection of a nonperfusion area caused by retinal vein occlusion in optical coherence tomography angiography images using deep learning.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets