Premium
Automatic identification of eyes at risk of developing idiopathic macular hole
Author(s) -
SILVA AS C.,
FIGUEIRA J,
LOBO C,
SIMÃO S,
MEIRELES A,
GOMES N,
BERNARDES R
Publication year - 2014
Publication title -
acta ophthalmologica
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.534
H-Index - 87
eISSN - 1755-3768
pISSN - 1755-375X
DOI - 10.1111/j.1755-3768.2014.3663.x
Subject(s) - macular hole , ophthalmology , medicine , internal limiting membrane , artificial intelligence , support vector machine , retinal , segmentation , limiting , optometry , computer science , visual acuity , mechanical engineering , engineering , vitrectomy
Abstract Purpose To automatically identify eyes with increased risk of developing idiopathic macular hole (IMH). This is proposed based on the fact that fellow eyes from patients with unilateral IMH have a risk increased by up to 29% of developing IMH. Our hypothesis is that there are changes in the topography of the macula that can allow the identification of these eyes prior to the development of IMH. Methods High‐resolution OCT scans from eyes at risk and control patients are acquired using the Cirrus HD‐OCT (Carl Zeiss Meditec, Dubin, CA, USA), Topcon 3D OCT (Topcon Medical Systems, Inc., Oakland, NJ, USA), and Spectralis HRA+OCT (Heidelberg Engineering GmbH, Heidelberg, Germany). From the volumetric scans, topographic maps are built from the segmentation of the inner limiting membrane and the retinal pigment epithelium. Parameters are extracted by fitting the maps to a set of bidimensional functions and from radial profiles of the region of interest. These are used to train a support vector machine (SVM), an automatic classification system. Optimizations to the system are ran in order to achieve the highest accuracy possible. Results Forty‐nine eyes with increased risk of IMH development (age: 67.4+‐5.5 yrs) and 44 control eyes (age: 58.5 +‐ 10.2) were imaged. The accuracy of the optimized SVM was 82.8%, sensitivity 85.7%, specificity 79.5%, assessed by 10‐fold cross‐validation. Conclusion The achieved results support our working hypothesis ‐ that structural changes are present well before being visible at a regular consultation.