Resumo

GIACINTI, David Jaime et al. Determination of the parabola of the retinal vasculature using a segmentation computational algorithm. Nova scientia [online]. 2019, vol.11, n.23, 00005.  Epub 10-Mar-2020. ISSN 2007-0705.  https://doi.org/10.21640/ns.v11i23.1902.

Quantitative analysis of the architecture of the superior and inferior temporal retinal veins and their monitoring over time could facilitate the diagnosis and timely treatment of diabetic retinopathy. This paper presents a novel method consisting of two stages for automatic segmentation and parabolic modeling of the superior and inferior temporal arcade vessels in retinal fundus images. In the first stage, the Multiscale Line Detector (MLD) is used to detect vessel-like structures in the retinal images. Since the MLD, is a vessel enhancement method, a thresholding strategy has to be used to classify vessel and non-vessel pixels, where an experimental threshold value is compared with five state-of-the-art thresholding methods. In this stage, the proposed segmentation method is compared with six state-of-the-art specialized methods in terms of segmentation accuracy. In the second stage, a parabolic modeling using an optimization strategy based on the Univariate Marginal Distribution Algorithm (UMDA) is performed over the segmented vessels and the results are compared with two state-of-the-art parametric methods and with the ground-truth images outlined by specialists. The results of vessel segmentation using the multiscale line detector demonstrated a high segmentation accuracy with a 0.9618 value using the DRIVE database of retinal fundus images. In addition, the parabolic modeling results provided an average accuracy of 0.825 with the ground-truth of the superior and inferior temporal arcade vessels outlined by ophthalmologists. According to the accuracy and the computational time (5.62 seconds) results, the proposed method can be considered as highly appropriate to perform computer-aided diagnosis in Ophthalmology.

Palavras-chave : estimation of distribution algorithms; computer-aided diagnosis; retinal fundus images; parabolic modeling; automatic segmentation.

        · resumo em Espanhol     · texto em Espanhol     · Espanhol ( pdf )