Fig 1.
Skeletonization and quantitative analysis of OCTA images.
We present an ImageJ macro that processes OCTA images of mCNV (A), (B). Our work enables users to perform quantitative analysis on skeletonized images (C), (D) of active mCNV lesions using nine computed biomarkers.
Table 1.
Inclusion and exclusion criteria for the mCNV study.
Fig 2.
The OCTA image processing pipeline and its various stages.
Our algorithm takes manually cropped OCTA images as input. The pipeline follows a branched structure. In Branch 1, we denoise the image (1A), apply the Frangi filter (1B), and use Median Thresholding (1C) to measure area-related biomarkers. In Branch 2, we apply the Mexican Hat filter (2A) and Binary Skeletonization (2B) to measure vascular biomarkers like junctions, diameter, fractal dimension, and tortuosity.
Fig 3.
Intermediate stages of the OCTA image as it travels down the pipeline.
(a) Input OCTA image showing the active mCNV lesion. (b) Manually delineated OCTA image. mCNV area is measured by counting the pixels within the contour. (c) The Gaussian kernel and Contrast adjustment were used for smoothing and denoising. (d) The Frangi vesselness filter and Mexican Hat filter were used to calculate Vessel Area and Vessel Density. (e) Binary skeletonization of the OCTA image using ImageJ. (f) The tagged skeleton was used to calculate the Junctions, Vessel Diameter, Tortuosity, and Fractal Dimension.
Table 2.
Statistical analysis of OCTA biomarkers on the 26-image dataset.
Table 3.
Comparison of our results (26 images) with a similar study on mCNV (Wang et al. [1], 31 images).