Dataset Information

Deep learning and ensemble stacking technique for differentiating polypoidal choroidal vasculopathy from neovascular age-related macular degeneration.

ABSTRACT: Polypoidal choroidal vasculopathy (PCV) and neovascular age-related macular degeneration (nAMD) share some similarity in clinical imaging manifestations. However, their disease entity and treatment strategy as well as visual outcomes are very different. To distinguish these two vision-threatening diseases is somewhat challenging but necessary. In this study, we propose a new artificial intelligence model using an ensemble stacking technique, which combines a color fundus photograph-based deep learning (DL) model and optical coherence tomography-based biomarkers, for differentiation of PCV from nAMD. Furthermore, we introduced multiple correspondence analysis, a method of transforming categorical data into principal components, to handle the dichotomous data for combining with another image DL system. This model achieved a robust performance with an accuracy, sensitivity, specificity, and area under the receiver operating characteristic curve of 83.67%, 80.76%, 84.72%, and 88.57%, respectively, by training nearly 700 active cases with suitable imaging quality and transfer learning architecture. This work could offer an alternative method of developing a multimodal DL model, improve its efficiency for distinguishing different diseases, and facilitate the broad application of medical engineering in a DL model design.

SUBMITTER: Chou YB

PROVIDER: S-EPMC8010118 | biostudies-literature |

REPOSITORIES: biostudies-literature

ACCESS DATA

Similar Datasets

Project description:PurposeA nonsynonymous coding variant in the manganese superoxide dismutase (SOD2) gene (V16A, rs4880) has been implicated in neovascular age-related macular degeneration (AMD). However, the findings have been inconsistent. Two studies in Japanese populations reported an opposite direction of association of the same allele at the V16A variant, whereas one study in a Northern Irish population found no effect of the variant on the risk of developing neovascular AMD. To address these apparently contradictory reports, we validated the association in a Japanese population.MethodsIn a Japanese population, we genotyped the V16A variant in 116 neovascular AMD patients, 140 polypoidal choroidal vasculopathy (PCV) patients, and 189 control participants. This association was also tested in a population of PCV participants to avoid variable findings across studies due to underlying sample heterogeneity and because disease phenotype was not well described in previous studies. We analyzed a tagging single nucleotide polymorphism (SNP) in addition to the V16A variant to capture all common SOD2 variations verified by the HapMap project. Genotyping was conducted using TaqMan technology. Associations were tested using single-SNP and haplotype analyses as well as a meta-analysis of the published literature. Population stratification was also evaluated in our study population.ResultsWe found no detectable association of the V16A variant or any other common SOD2 variation with either neovascular AMD or PCV, as demonstrated by both single-SNP and haplotype analyses. Population structure analyses precluded stratification artifacts in our study cohort. A meta-analysis of the association between the V16A variant and neovascular AMD also failed to detect a significant association.ConclusionsWe found no evidence to support the role of any common SOD2 variations including the V16A variant in the susceptibility to neovascular AMD or PCV. Our study highlights the importance and difficulty in replicating genetic association studies of complex human diseases.

Project description:BackgroundTo compare the choriocapillary flow density (CFD) among the fellow eyes of polypoidal choroidal vasculopathy (PCV), neovascular age-related macular degeneration (nAMD), and healthy controls using spectral-domain optical coherence angiography tomography (SD-OCTA).MethodsThis is a cross-sectional study that includes the fellow eyes of 38 patients with unilateral PCV, 36 patients with unilateral nAMD, and 36 eyes from 36 healthy volunteers. The PCV group was further classified into polypoidal CNV (P-CNV) and typical PCV (T-PCV) for subgroup analysis. The age, subfoveal choroidal thickness (SFCT), Age-Related Eye Disease Study (AREDS) classification, and fellow eye diagnosis were acquired. All subjects underwent SD-OCTA with a 6.0-mm scan pattern. Circles with radius of 1.00, 1.50, and 3.00?mm were manually selected in the choriocapillaris (CC) slab, and the CFD was calculated as the percentage of the flow area to the whole selected area as CFD-1.00, 1.50, and 3.00, respectively. Univariate and multivariate analysis were performed to study the correlation between the aforementioned factors with CFD.ResultsThe mean CFD-1.00, 1.50, and 3.00 of the nAMD group were 61.51, 63.18, and 66.20, respectively; these were significantly lower than those of the PCV group (65.90, 66.89, and 67.94; P?<?0.001, P?<?0.001, and P?=?0.010; respectively) and control group (66.28, 66.96, and 68.42; P?<?0.001, P?<?0.001, and P?=?0.001, respectively), and no difference was detected between the PCV and control group or between PCV subtypes. The AREDS classification and fellow eye diagnosis were correlated with CFD in univariate analysis; however, only the fellow eye diagnosis showed a significant correlation after multiple linear regression.ConclusionsThe CFD of nAMD fellow eyes was significantly lower than that of PCV and control eyes, and no difference was detected between PCV and control group, indicating that CC loss plays a different role in the early pathogenesis of nAMD and PCV.

Project description:PURPOSE: Using a candidate-gene approach, a recent case-control study identified a previously unknown association between neovascular age-related macular degeneration (AMD) and the coding Met72Thr variant in the pigment epithelium-derived factor (PEDF) gene in a Taiwan Chinese population. However, a subsequent replication study failed to see this association in a white European population. We noted an important difference in the sample ascertainment scheme between these two studies. The original study did not consider findings of indocyanine green (ICG) angiography for disease classification, which is the only way to obtain a clear image of polypoidal choroidal vasculopathy (PCV) lesions. This suggests that their cohort might include a considerable amount of PCV, given its high prevalence in the Chinese population. In contrast, the replication study intentionally excluded PCV from the case cohort on the basis of ICG angiograms. Therefore, the inconsistent finding might be caused by potential sample heterogeneity between these two studies. In this respect, this association needed to be examined in a case series of clearly defined individuals with neovascular AMD and PCV. The aim of this study was to validate the previously reported association of the PEDF Met72Thr variant in a well characterized Japanese population with neovascular AMD and PCV. METHODS: We genotyped the Met72Thr variant (rs1136287) in 116 patients with neovascular AMD, 140 patients with PCV, and 189 control participants in a Japanese population. Genotyping was performed using TaqMan technology. We tested for an association of this variant with neovascular AMD and PCV separately. We also evaluated population stratification in our study cohort. RESULTS: We found no statistically significant evidence for association between rs1136287 and either neovascular AMD or PCV under any genetic models (trend, genotypic, dominant, and recessive genetic models; p>0.05). Population structure analyses excluded stratification artifact in our study population. CONCLUSIONS: We report a lack of association between the PEDF Met72Thr variant and either neovascular AMD or PCV in a Japanese population. We conclude that the Met72Thr variant does not play a significant role in the risk of developing neovascular AMD or PCV.

Project description:Background: Neovascular age-related macular degeneration (nAMD) and polypoidal choroidal vasculopathy (PCV) are major causes of blindness in aged people. 30% of the patients show unsatisfactory response to anti-vascular endothelial growth factor (anti-VEGF) drugs. This study aims to investigate the relationship between serum metabolome and treatment response to anti-VEGF therapy. Methods: A prospective longitudinal study was conducted between March 2017 and April 2019 in 13 clinical sites in China. The discovery group were enrolled from Shanghai General Hospital. The validation group consisted of patients from the other 12 sites. Participants received at least one intravitreal injection of 0.5 mg anti-VEGF drug, conbercept, and were divided into two groups - responders and non-responders. Serum samples of both groups were processed for UHPLC-MS/MS analysis. We constructed principal component analysis (PCA) and partial least squares discriminant analysis (PLS-DA) models to investigate the metabolic differences between two groups using SIMCA-P. Area under curve (AUC) was calculated to screen the biomarkers to predict treatment response. Metabolites sub-classes and enriched pathways were obtained using MetaboAnalyst5.0. Results: 219 eyes from 219 patients (nAMD = 126; PCV = 93) were enrolled. A total of 248 metabolites were detected. PCA and PLS-DA models of the discovery group demonstrated that the metabolic profiles of responders and non-responders clearly differed. Eighty-five differential metabolites were identified, including sub-classes of diacylglycerophosphocholines, lysophosphatidylcholine (LPC), fatty acids, phosphocholine, etc. Responders and non-responders differed most significantly in metabolism of LPC (p = 7.16 × 10^-19) and diacylglycerophosphocholine (p = 6.96 × 10^-17). LPC 18:0 exhibited the highest AUC, which is 0.896 with 95% confidence internal between 0.833 and 0.949, to discriminate responders. The predictive accuracy of LPC 18:0 was 72.4% in the validation group. Conclusions: This study suggests that differential metabolites may be useful for guiding treatment options for nAMD and PCV. Metabolism of LPC and diacylglycerophosphocholine were found to affect response to conbercept treatment. LPC 18:0 was a potential biomarker to discriminate responders from non-responders.

Dataset Information

Deep learning and ensemble stacking technique for differentiating polypoidal choroidal vasculopathy from neovascular age-related macular degeneration.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets