Purchase this article with an account.
Pedro Guimaraes, Luisa Frizziero, Jeffrey C Wigdahl, Edoardo Midena, Alfredo Ruggeri; Fully-automatic segmentation of conjunctival blood vessels. Invest. Ophthalmol. Vis. Sci. 2015;56(7 ):1970.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
The conjunctiva is one of the few places in the human body where one can easily and directly observe the microvasculature. In this work, we propose a novel fully-automatic approach to segment the conjunctival blood vessels.
Conjunctival blood vessels can be imaged noninvasively using slit-lamp microscopy. In this work, the conjunctiva of 10 patients was imaged using a digital camera attached to a SL 990 slit lamp microscope (C.S.O., Scandicci, Italy). For each patient, 4 images of 960x1280 pixels with a magnification of 10x were acquired - one image per region (inferior, superior, nasal, and temporal).<br /> When imaging the conjunctiva, other structures such as the iris or eyelashes may also be present in the image. For this reason, a region of interest (ROI) is automatically defined, where the segmentation is performed. As a preprocessing step, each image undergoes top-hat filtering. This method has been shown to correct for different imaging artifacts such as uneven illumination and contrast.<br /> Vessel segmentation is achieved using the phase-congruency parameter, a dimensionless quantity computed for each image pixel and invariant to contrast and scale. It measures the agreement of the phase of the Fourier components of an image, with the same relevance being given to all frequency components, independent of gradient magnitude. The phase congruency is thresholded to obtain a binary image. Finally, fragmented vessel segments are connected by morphological dilation of the resulting binary images, while isolated small segments are erased.
All images could be segmented with the proposed algorithm. Figure 1 shows two examples of the results, where it appears that most blood vessels are correctly traced. The time required to analyze a single image, using a prototype developed in MATLAB language, was 1.83 ± 0.04 s (mean ± standard deviation).
The proposed algorithm seems capable of correctly tracing the conjunctival blood vessels. This approach opens the way to fully-automatic analysis of microvasculature, which may provide important clinical insight for several pathologies.
This PDF is available to Subscribers Only