Quantificação de angiogênese corneana in vivo através de processamento de imagens digitais

Abstract

A software was developed specifically for segmentation and quantification of corneal angiogenesis in digital images acquired in vivo. The purpose was to establish an automatic method to quantify the corneal progression of the neovascular response in a rapid, objective and accurate manner. It was also developed a model of inflammatory corneal angiogenesis in rabbits induced by a punctual alkaline cauterization. Angiogenic response was evaluated at days 3, 6, 9, 12, 15, 18 and 21 after cauterization. At these occasions, digital images of cornea were also captured in a standardized fashion. The software is composed of five modules, each with specific functions, which automatically identify and segment the neovessels and quantify the neovascular response through the calculation of three main parameters: neovascularization area, total vascular length and blood vessels number. Thus, the system does not require previous enhancement of the neovascularization, so that it provides the temporal and spatial monitoring of the angiogenic response in vivo. To assess the validity of the automatic method, it was determined the correlation degree between each automatic parameters and the five manual variables. It was also evaluated by comparing automatic recognition of blood vessels performed by the software on 50 subimages with the manual recognition performed by three raters by consensus (gold-standard) and then calculating the sensitivity, specificity and kappa coefficient. To assess reliability, three raters used the automatic method for analysing 50 images at two moments. The intra and inter-raters agreement was assessed by calculating intraclass correlation coefficient. To illustrate the utility of the automatic method, it was designed an assay that evaluated the effect of prednisone, thalidomide, acetylsalicylic acid, etoricoxib and celecoxib in the model of inflammatory corneal angiogenesis. The analysis of the graphics of the manual and automatic parameters showed that the temporal progression of the neovascular response in this model followed a biphasic pattern: exponential growth until day 12 (proliferation phase), followed by linear growth until day 21 (maturation phase). It was found a statistically significant positive linear correlation between automatic and manual variables, denoting that both methods measured the angiogenic response in an analogous fashion. The sensitivity of the automatic method was 85.32% and the specificity was 96.37%. The agreement between manual and automatic recognition of blood vessels was classified as almost perfect, according to kappa statistics. The inter-raters agreement was measured in six situations and it was classified as good in two of them and as excellent in the others. Among the nine situations analysed, the intra-rater agreement was classified as good in only one of them and as excellent in the others. Furthermore, the automatic system was able to detect differences in the intensity of the antiangiogenic effect of tested drugs, allowing the inhibitory activity to be graduated as full (prednisone), partial (thalidomide, acetylsalicylic acid and etoricoxib) and insufficient (celecoxib), demonstrating, thus, its value as method for quantitative analysis of angiogenesis. Therefore, such results demonstrate that the proposed automatic system constitutes an accurate and reproducible method of angiogenesis quantification.