Purchase this article with an account.
Joao Figueira, Helena Pereira, Mario Alfaiate, Ana R. Santos, Torcato Santos, L Pedrosa, M Mendes, R Venancio, Rui Bernardes; Identification of eyes at risk of developing Idiopathic Macular Holes by Support Vector Machines. Invest. Ophthalmol. Vis. Sci. 2012;53(14):5219.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To discriminate between healthy eyes and eyes at risk of developing Idiopathic Macular Holes (IMH).
Eighteen eyes from 10 healthy volunteers, aged 58.24 ± 13.36 (m ± SD) (3 male and 7 female), and 18 fellow eyes from 18 patients with unilateral IMH, (67.33 ± 8.64 years) (10 male and 8 female) were imaged by Cirrus HD-OCT (Carl Zeiss Meditec, Dublin, CA, USA) in order to compare their foveal areas. Eyes with glaucoma, myopia (> 3D) or other retinal pathology were excluded from the present study. The shape of the inner limiting membrane was depth-wise normalized by the retinal pigment epithelium to obtain the depth-wise corrected shape of the retinal surface. In order to describe the shape of the retinal surface with special focus on the foveal depression, fits of well known mathematical functions were used to model the three-dimensional shape of the region of interest. The set of fits performed allows to synthesize each individual retina shape in a set of parameters describing each of mathematical function used. The working hypothesis is that differences can be found within this set of functions' parameters between the two groups of eyes. We resort to the pattern classification support vector machine (SVM) algorithm to discriminate between groups through the training using known cases. The assessment of performance of the SVM was therefore tested using the leave-one-out approach due to the small number of cases available, i. e., the system is trained using all but one case each time, which is then tested to be classified into one of the two possible groups.
Eleven (61.11%) of the 18 fellow IMH eyes were correctly classified in the fellow IMH group, and 7 (38,89%) were classified in the control group. Four (22.22%) of the 18 control eyes were classified in the fellow IMH group, and 14 (77,78%) were correctly classified in the control group. With this technique, 25 (69%) of all eyes were correctly classified. Although the number of cases available is relatively small, the findings allowed us to conclude on the possibility to discriminate between healthy eyes and eyes at risk of developing IMH. Further analyses need to be carried out using a larger population since the number of parameters herewith used exceeds by far the number of cases. Nevertheless, this proof-of-concept suggests the feasibility of the proposed methodology.
Using a noninvasive imaging technique able to map the shape of the foveal/parafoveal area and a fully automatic classification system, we demonstrated the possibility to correctly classify eyes in the group of healthy volunteers or in the group of eyes at risk of developing IMH in over 60% of the case (69% of the cases overall).
This PDF is available to Subscribers Only