Early detection of retinopathy in the periphery of the macula is an important step in preventing severe vision loss. Some morphological parameters about the extensive retina can be obtained through ultra-wide-field OCTA images. Based on small-scale fundus OCTA vessel segmentation, accurate diagnosis can already be obtained by means of deep learning. However, no similar research of segmentation of peripheral blood vessels is reported. In this study, blood vessels of retina were segmented, and blood vessel centerlines were extracted in ultra-wide-field OCTA images. Quantification of the segmented images was performed to explore features of blood vessel. We used a U-shaped neural network that performs well on small samples to cope with the problem of limited data sets. Scale compression and slice segmentation were used to apply the trained network model to vessel segmentation and centerline extraction in ultra-wide-field OCTA images which is of size at 21mm×21mm. Based on the results of the segmentation of blood vessels, the diameter index of blood vessels and vascular tortuousness were calculated, which proved to be associated with some eye diseases. These results and parameters can be helpful for the early screening of some ophthalmic diseases.
Diabetic retinopathy (DR) accounts for accumulated damage to retinal blood vessels which can lead to blindness if it is not detected in its early stage. Optical coherence tomography angiography (OCTA) provides noninvasive and dye-free method to assess 3D retinal and choroid circulations which has been used to evaluate DR ever since it was proposed. In this study, widefield OCTA (WF-OCTA) images were provided by the swept-source optical coherence tomography (SS-OCT) with a 12mm×12mm single scan centered on the fovea and a convolutional neural network (CNN) model was proposed to extract small lesions present in images for the early detection of DR. The proposed model achieved a classification accuracy of 95%, sensitivity of 97.12% and specificity of 87.90% in detecting DR. The accuracy of the model for DR staging is 85.74%, which is higher than that of the Vgg16 by 5.76% and the Inception-V3 by 4.49%. This work demonstrated reproducible and consistent detection results with high sensitivity and specificity.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.