| Part ? Exploring the effect of hypertension on retinal microvasculature based on retinal fundus photographs using deep learningBackground: The phenotypic information of retinal microvasculature which is the part of cardiovascular system can be viewed directly and noninvasively and analyzed qualitatively and quantitatively.So a lot of research tried to establish the connection between microvascular abnormalities in fundus images and systemic vascular pathologic changes and retinal microvascular abnormal signs were used as a potential early warning signal of cardiovascular disease.Objective: Hypertension is the leading risk factor of cardiovascular disease and has profound effects on both the structure and function of the microvasculature.This study aimed to use deep learning technique that can detect subclinical features appearing below the threshold of a human observer to explore the effect of hypertension on morphological features of retinal microvasculature.Materials and methods: We collected 735 patients(1007 eyes)with a diagnosis of hypertension and 684 normotensive control subjects(1005 eyes),who were admitted between May 2017 and December 2018 to Shenyang He Eye Hospital with eye disease.Retinal photographs that were centered on the macula and documented the optic disc,the macula,substantial portions of the temporal vascular arcades were taken with 45°digital camera.To highlight the retinal vasculature,two image enhancement methods,gamma correction and contrast limited adaptive histogram equalization(CLAHE),were conducted to enhance the image contrast,and “enhanced dataset” was obtained.The “enhanced dataset” consisted of 2012 fundus images,of which 1007 were hypertension images and 1005 were non-hypertension images.We used the trained U-net to extract retinal vessels of every fundus image of “enhanced dataset”,and “segmented dataset” was obtained.All fundus images have one target label: hypertension-status(Yes/No),1 represents hypertension images and 0 represents non-hypertension images.Respectively using “enhanced dataset” and “segmented dataset”,we trained small convolutional neural networks(CNN).All fundus images were randomly partitioned into five equal sized subsamples.Of the five subsamples,a single subsample was retained as the “test set” which was not used during the training process to test the model,and the remaining four subsamples are used as the “development set” to develop our model.The training process was then repeated five times.To evaluate performances of the learned model,we adopted several evaluation parameters,including accuracy,specificity,precision,recall,and the area under the receiver operating characteristic curve(AUC).Using “segmented dataset”,we used a deep learning technique called Gradient-weighted Class Activation Mapping(Grad-CAM)to generate heat maps for the class “hypertension” and “non-hypertension”.Results: For “enhanced dataset”,averaged the results of the five cross-validation and our model achieved an accuracy of 56.76%,a specificity of 63.80%,a precision of 58.97%,and a recall of 49.93%.The AUC was 0.6069.For “segmented dataset”,our model produced an improved accuracy 60.94% and recall 70.48%,a similar precision 59.27%,but it's specificity dropped to 51.54%.Finally,the AUC was 0.6506.For heat maps for the class “hypertension”,red patchy areas that were strongly activated showed discrete distribution,and most of them were on or around arterial/venous bifurcations.But,for heat maps for the class “non-hypertension”,red areas showed a continuous distribution along the blood vessels.Conclusions: The change of the bifurcation pattern of retinal microvasculature was probably the most significant in response to elevated blood pressure.Part ? Exploring the effect of diabetes on retinal microvasculature based on retinal fundus photographs using deep learningBackground: The phenotypic information of retinal microvasculature which is the part of cardiovascular system can be viewed directly and noninvasively and analyzed qualitatively and quantitatively.So a lot of research tried to establish the connection between microvascular abnormalities in fundus images and systemic vascular pathologic changes and retinal microvascular abnormal signs were used as a potential early warning signal of cardiovascular disease.Objective: Diabetes is the leading risk factor of cardiovascular disease and has profound effects on both the structure and function of the microvasculature.This study aimed to use deep learning technique that can detect subclinical features appearing below the threshold of a human observer to explore the effect of diabetes on morphological features of retinal microvasculature.Materials and methods: We collected 479 patients(534 eyes)with a diagnosis of diabetes and 375 non-diabetic control subjects(550 eyes),who were admitted between May 2017 and December 2018 to Shenyang He Eye Hospital with eye disease.Retinal photographs that were centered on the macula and documented the optic disc,the macula,substantial portions of the temporal vascular arcades were taken with 45°digital camera.To highlight the retinal vasculature,two image enhancement methods,gamma correction and contrast limited adaptive histogram equalization(CLAHE),were conducted to enhance the image contrast,and “enhanced dataset” was obtained.The “enhanced dataset” consisted of 1084 fundus images,of which 534 were diabetes images and 550 were non-diabetes images.We used the trained U-net to extract retinal vessels of every fundus image of “enhanced dataset”,and “segmented dataset” was obtained.All fundus images have one target label: diabetes-status(Yes/No),1 represents diabetes images and 0 represents non-diabetes images.Respectively using “enhanced dataset” and “segmented dataset”,we trained small convolutional neural networks(CNN).All fundus images were randomly partitioned into five equal sized subsamples.Of the five subsamples,a single subsample was retained as the “test set” which was not used during the training process to test the model,and the remaining four subsamples are used as the “development set” to develop our model.The training process was then repeated five times.To evaluate performances of the learned model,we adopted several evaluation parameters,including accuracy,specificity,precision,recall,and the area under the receiver operating characteristic curve(AUC).Using “enhanced dataset”,we used a deep learning technique called Gradient-weighted Class Activation Mapping(Grad-CAM)to generate heat maps for the class “diabetes”.Using “segmented dataset”,we used Grad-CAM to generate heat maps for the class “diabetes” and “non-diabetes”.Results: For “segmented dataset”,averaged the results of the five cross-validation and our model achieved an accuracy of 65.78%,a specificity of 71.12%,a precision of 66.52%,and a recall of 60.22%.The AUC was 0.7003.For “enhanced dataset”,all evaluation indexes increased significantly.The model achieved an accuracy of 74.82%,a specificity of 77.62%,a precision of 75.69%,and a recall of 72.04%.The AUC was 0.8324.Using “enhanced dataset”,the heat maps for the class “diabetes” showed that red areas were mainly distributed in four areas(macular area,the area around the optic disc,retinopathy area and microvascular area).Using “segmented dataset”,the heat maps for the class “diabetes” showed that red areas were mainly distributed in three areas: the patchy red areas were distributed on or around arterial/venous bifurcations,the strip red areas were distributed in the retinal arteriole,the strip red areas were distributed in the retinal venule.But,for heat maps for the class “non-diabetes”,red areas showed a continuous distribution along the blood vessels.Conclusions: Compared with non-diabetic fundus image,the main morphological features of diabetic fundus image were distributed in four areas: macular area,the area around the optic disc,retinopathy area and microvascular area with abnormal morphology.The morphological changes of retinal microvasculature under the condition of diabetes mainly include: the change of bifurcation pattern of retinal microvasculature,the morphological change of retinal arteriole and venule. |
PDF Full Text Request