3D shape representation of OCT Angiography images

Abstract

Purpose : The studies of microvascular structures in 2D enface projection of 3D OCTA images recently show great promise in retinal image analysis field. However, direct analysis of 3D OCTA images which have rich depth-resolved microvascular information are rather limited due to low vessel visibility and strong projection artifacts. We aim to establish a 3D data representation of OCTA images via advanced shape modeling techniques to provide a high-quality microvascular visualization and to preserve 3D geometric and topological information for further analysis.

Methods : We propose a novel framework including denoising, vessel enhancement, segmentation and triangular mesh representation to process 3D microvasculature in OCTA images. 3D curvelets denoising is employed to suppress speckle noise and preserve elongated structures for better continuity. The optimally oriented flux filter is applied to enhance vessels with best adaption to local scales. A binary vessel network is obtained by thresholding the enhanced OCTA images. Then, we take advantage of the established shape modeling technique to reconstruct a 3D vessel triangular mesh representation. A dataset of 200 repeated scans from 20 eyes was available for validation.

Results : The proposed pipeline has been evaluated on the 3D OCTA repeated scans to verify its repeatability. Fig. 1 shows a typical example of 3D reconstructed vessel triangular meshes of a candidate scan and its overlapping with the template mesh surface. The average distance calculated from the 200 scans is 0.0056(mm), much smaller than the a voxel diagonal length 0.0174(mm). A high mean similarity ratio of 0.9324 has been obtained between scans of each subject. The high performance demonstrates an effective shape representation framework with strong repeatability.

Conclusions : The proposed framework provides an intuitive 3D mesh representation of OCTA images. Geometrical and topological analysis on 3D vessel meshes can be further exploited for describing disease related pathology.

This abstract was presented at the 2019 ARVO Imaging in the Eye Conference, held in Vancouver, Canada, April 26-27, 2019.