Automatic Analysis Of Fudus Blood And Its Relationship With Adolescent Myopia

Part ⅠBackground: Choroid vascular layer is crucial for the oxygenation and metabolism activity of outer retina of the human eyes.Advent noninvasive imaging modality of swept source optical coherence tomography（SS-OCT）is now enabling the depiction of detailed structures of the choroidal vessels.Choroidal vessels’ image on SS-OCT were great supplement to current imaging modalities focusing choroid.However,the manual segmentation of the choroidal vessels is a time consuming,nonreproducible and error-prone process.Existing OCT based segmentation methods are unreliable for the challenging choroidal vessels segmentation,which exhibit fuzzy boundaries,low contrast with the background and inhomogeneous textures.Purpose: To develop a automatic algorithm for the identification and segmentation of choroidal vessels on SS-OCT imagesMethods: To overcome these limitations,in this work,we propose to leverage a Refine Net architecture to segment choroidal vessels.Refine Net architecture is a deep neural network that achieves segmentation by exploiting and hierarchically combining features at multiple levels in a cascaded manner.We evaluated the proposed method with a dataset contained 40 pairs of choroid images,approximately 3600 annotated regions,which were manually annotated by two clinicians independently.A 5-folds cross-validation approach was used for comparison and a segmentation agreement calculation approach was used to compare the variability between auto-segmentation,clinician 1（C1）and clinician 2（C2）.We further compared with another recently proposed deep neural network based segmentation approach-fully convolutional networks（FCN）,which has been widely applied for segmentation on OCT images.Results: For the proposed method,the segmentation agreement（SA） between the segmentation from Refine Net and clinician 1（C1）was0.840±0.0354 and between the Refine Net and clinician 2（C2）was0.823±0.0265,which are both higher than between the C1 and C2 of0.821±0.0370.The proposed method also outperformed segmentation of FCN,whose SA to C1 of 0.780±0.0566 and SA to C2 of 0.785±0.0436,with statistic significance（P<0.001,paired t-test）.The intraclass correlation coefficient（ICC）and Bland-Altman analysis showed good reproducibility.Conclusion: The proposed algorithm was capable of segmenting the choroidal vessels in SS-OCT,providing an accurate and automated approach to quantitatively analyze of choroidal vessels and we suggest this will permit new researches and clinical analyses related to choroid.Part Ⅱ Purpose: To investigate the difference of choroidal vessels on SS-OCT images between high myopia and normal adolescents.Methods: The Refine Net deep neural network model was used to automatically identify the choroidal vessels in the SS-OCT image,and the results were analyzed.The total choroidal volume,the volume of choroidal vessels and the density of the choroidal vessels were calculated respectively in the high myopia group and the normal group.Results: On each B-scan,the area of choroidal vessels in the high myopia group was 0.797± 0.270 mm2,which was significantly lower than that in the normal group,1.325±0.361（p<0.001）.The total choroidal area of the high myopia group was 2.261±0.505 mm2,which was lower than that in the normal group,3.196±0.599（p<0.001）.The average ratio between vessel to choroid in high myopia group was 0.344±0.055,which was also lower than that in the normal group,0.408±0.053（p<0.001）.Conclusions: In the early stage of adolescence,the total volume of choroid is smaller than that of normal adolescents at the early stage of high myopia.The volume of choroidal vessels is smaller than that of normal adolescents,and the density of choroidal vessels is smaller than that of normal adolescents.Part Ⅲ Purpose: The aim of the present study was to investigate the age-specific normative values of retinal oxygen saturation and explore their associated factors in healthy Chinese school-aged children with different refractive statuses.Methods: A population-based observational cross-sectional study Children aged 7 to 19 years were enrolled.Each participant underwent a series of comprehensive ocular examinations,including axial length（AL）,cycloplegic-refraction and Oxymap T1 images taken after cycloplegia.The acquired oximetry images were measured,and the values of the retinal oxygen saturation parameters were calculated.The independent factors of the retinal oxygen saturation were analyzed using multiple linear regression.The oxygen saturation of retinal arteries（Sa O₂）and veins（Sv O₂）,and the difference between arteries and veins（AVD）,were measured as the main outcomes.Results: In all,1461 participants were included with a mean age of 12.1±3.2 years,and 53.0% were boys.The mean Sa O₂,Sv O₂ and AVD were 83.7±6.4%,50.1±5.4%,33.6±5.4%,respectively,and the values increased with age.Girls had higher Sv O₂ and lower AVD than boys（p<0.05）.The Pearson correlation coefficients among Spherical equivalent（SE）and Sa O₂,Sv O₂,and AVD were-0.372,-0.203,and-0.240,respectively（all p<0.001）,while the correlations between AL and Sa O₂,Sv O₂,and AVD were 0.276,0.106,and 0.221,respectively（all p<0.001）.The myopia group had significantly higher Sa O₂,Sv O₂ and AVD than the emmetropia and hyperopia groups（p<0.001）,but the high myopia group had lower Sa O₂ and Sv O₂ than moderate myopes.When age,gender,SE,BMI,and IOP were included as factors in the multiple regression,it demonstrated that older age and more myopic SE were commonly associated with higher Sa O₂,Sv O₂ and AVD.Gender was an independent factor predicting Sv O₂,while gender and BMI were the independent factors predicting AVD.Age explained more variance than SE in Sa O₂ and Sv O₂,while for AVD,SE explained the most variance,followed by age.Conclusions: Myopic refractive errors and eye axials increased with Sa O₂,Sv O₂ and AVD.In the condition of mild and moderate myopia,the retina has certain compensatory mechanism to improve the oxygen saturation of the retinal arteries.While In the high myopia status,this compensatory ability reaches its limit,and therefore,it becomes a decompensated state.The decompensation plays an important role in the formation and development of high myopia retinopathy...