Purchase this article with an account.
Ana Sílvia C. Silva, João Figueira, Sílvia Simão, Nuno Gomes, Carlos Neves, Angelina M Silva, Natália Ferreira, Rui Bernardes; Preclinical Identification of Eyes at risk of Developing Idiopathic Macular Hole — An Update. Invest. Ophthalmol. Vis. Sci. 2014;55(13):324.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To discriminate healthy eyes from eyes with increased risk of developing idiopathic macular holes (IMH).
Although causes for IMH aren't fully understood, it is well known that patients with unilateral IMH have increased probability of developing IMH in their other eye (a risk increased by 10-20%). In this work, we present further evidence to the hypothesis presented in 2012 (Invest Ophthalmol Vis Sci 2012;53: E-Abstract 5219): that it is possible to detect this increased risk of developing IMH based on information present in an optical coherence tomography. Using Cirrus HD-OCT (Carl Zeiss Meditec, Dublin, CA, USA), we imaged the macular region of eyes at risk of developing IMH and eyes from a control group. In neither group were included eyes with glaucoma, myopia (> 3D) or other retinal pathology. To obtain a map of the retinal surface, the shape of the inner limiting membrane was depth-wise corrected by the retinal pigment epithelium. The surface shape is described by a set of parameters obtained by fitting a set of well-known 2D mathematical functions to the retinal maps. In addition, features from a set of bidimensional profiles are added, to improve discrimination, namely the curvature, angle at the fovea, and the volume of the foveal depression. A total of 27 parameters are thus computed per eye. Twenty-four eyes from 24 patients with unilateral IMH (18 women, age 67.2±6.8 years), and 32 eyes from 19 patients (13 women, age 59.3±7.3 years) were imaged and processed. We resort to support vector machines (SVM), a supervised learning model, to classify eyes into the group of healthy and at risk. To improve the performance of the classification, backward elimination and forward selection routines were ran to identify an optimal set of features. We used an N-fold cross-validation process (N=10) to determine the system performance.
For a reduced set of features we were able to achieve an accuracy of 83.6%, sensitivity of 66.7% and specificity of 96.8%. These results represent a step forward as compared to the ones previously presented by our group.
The results here achieved not only reinforce the previously demonstrated possibility to identify eyes at risk of developing IMH using noninvasive imaging techniques, but also show that the improvements made over the last years are yielding positive results, demonstrating an increase in the reliability of our test.
This PDF is available to Subscribers Only