Blood Vessel Segmentation and 3D Reconstruction of Neovascularization in Age-Related Macular Degeneration from OCTA Scans
摘要
We propose here a novel method for vessel segmentation and 3D reconstruction of nAMD from OCTA images. The method has four steps: (1) creating vessel mask; (2) isolating the Macular Neovascular Membranes from noise; (3) finding the vessel depth in volume; (4) reconstructing the 3D volume. Blood vessels were identified by detecting and marking spikes from four different directions—row, column, and two diagonals—treated as a wave form, combined with cross-correlation among five neighbouring points. When the parameters are manually optimised for the available dataset, the method can achieve up to 100% accuracy. Furthermore, using a kernel extracted from the main image, the depth of a blood vessel was determined with an accuracy of 95.6%. By extracting a 3D volume centred around the point of interest, based on previously calculated coordinates, and retaining only high-intensity points, a 3D reconstruction of the neovascularization area in AMD pathology was achieved. The method has been tested on a dataset provided by the Emergency County Hospital, Cluj-Napoca, Romania.