Purchase this article with an account.
D. Cabrera, R.W. Knighton; Active Contour Models for Assessing Lesion Shape and Area in OCT Images of the Retina . Invest. Ophthalmol. Vis. Sci. 2003;44(13):1770.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Purpose: To evaluate the ability of an active contour model to yield accurate shape descriptions of retinal lesions associated with neovascular age-related macular degeneration. Methods: An active contour model (GVF snake) that uses a gradient vector flow field as its external force was applied to OCT images of retinas demonstrating cystoid and subretinal fluid spaces. Several implementation issues were addressed, such as the initialization conditions, the parameter values, the number of iterations, the values of the regularization parameter, and the computation of the edge maps. Nonlinear anisotropic diffusion filtering and a gaussian convolution operator were explored to control the effect of spurious edges inside the retinal lesions on the motion of the contour toward the desired edges. Once the contours of the lesions were outlined, quantitative analysis of the surface area of the lesions across the OCT radial scans was performed. Results: The nonlinear anisotropic diffusion filter can be used as a preprocessing step to an intensity-based segmentation technique, resulting in images simpler to segment than the originals, especially in the presence of speckle noise. The active contour algorithm improves when the image contains little noise and soft, but obvious edges. The improvement comes from the fact that, in such an image, the contour is unlikely to attach itself to noise and is more likely to find the wider edges obtained after the anisotropic diffusion process. The GVF snake model is sensitive to initialization (position, size and shape of the initial snake); and it is very sensitive to changes of the regularization parameter values. Nevertheless, the GVF snake model could accurately outline fluid-filled lesions within and under the retina. Conclusions: The detection method tested proved effective in capturing the complexity of lesions in OCT images. Further investigation is required in order to optimally choose the regularization parameter according to the amount of noise present in the image. Active contour models combined with nonlinear anisotropic diffusion filtering show promise in the detection of retinal features in clinical OCT images. Since the number and size of lesions indicate the progression of the disease in the patient, lesion detection may significantly aid in analysis of treatments and diagnosis.
This PDF is available to Subscribers Only