Abstract
Purpose :
OCTA analysis typically involves 2D quantification based on the enface representation of retinal layers. 2D quantitative metrics can obscure the geometric and morphological information in 3D vasculature. There is a need for comprehensive 3D quantitative tools to more precisely model the complex dimensions of retinal vasculature available in OCTA scans. The aim of this study is to provide a robust and flexible 3D framework for the analysis of OCTA data.
Methods :
We propose an automated 3D framework for an end-to-end analysis of OCTA images. Initially, OCTA retina region extracted using OCT explorer software. In the second step, 3D Fast Discrete Curvelet Transforms was used as a denoising and enhancement tool. Optimally Oriented Flux (OOF) was applied in a hybrid setting with curvelet to generate a clear vessel map. Binary vessel segmentation of OCTA volumes were then obtained by thresholding on OOF response. Ultimately, a 3D vessel skeleton was calculated by incorporating the Hamilton-Jacobian method. The proposed method is very flexible with respect to the selection of OCT layers and is not limited to only superficial or deep retina layers. Moreover, this method provides 3D analysis and quantification of all retina vessels and small capillaries that has the potential to be valuable in early diagnoses of DR.
Results :
We validated our method on the OCTA images of a large-scale (n=338) DR study. Fig.1 demonstrates one representative eye selected from each group: Normal, Mild, Moderate, Severe, and PDR. For the quantitative assessment, 3D small vessel skeleton density (SVSD) of all eyes was extracted. The result of statistical analysis of healthy control compared to each stage of DR shows significant difference of SVSD in superficial and deep layers (p< 0.05). The test also reveals more significant difference (p< 0.001) in superficial layers for for earlier stage of DR (Mild), and our sublayer analysis localize the superficial layer alternation mainly in RNFL-GCL layer.
Conclusions :
In this work we provide a robust, flexible and automatic 3D framework for analysis of retinal vasculature available in OCTA scans. The proposed method was tested on a large-scale Diabetic Retinopathy dataset using small vessel skeleton density (SVSD) feature. The quantitative assessment shows significant difference of SVSD in superficial and deep layers of each stage of DR compared to healthy control.
This abstract was presented at the 2019 ARVO Imaging in the Eye Conference, held in Vancouver, Canada, April 26-27, 2019.