Purchase this article with an account.
M. Wang, A. Lu, D. Huang, Advanced Imaging for Glaucoma Study Group; Combining Information From Three Anatomic Regions in the Diagnosis of Glaucoma With Optical Coherence Tomography. Invest. Ophthalmol. Vis. Sci. 2008;49(13):3640. doi: https://doi.org/.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To combine optical coherence tomography (OCT) measurements of the optic nerve head (ONH), circumpapillary nerve fiber layer (cpNFL) and macular thickness (MT) in the diagnosis of glaucoma.
Eighty-nine age-matched normal (N) and perimetric glaucoma (PG) participants were randomly selected from the Advanced Imaging for Glaucoma Study database. OCT scanning was performed with the Stratus software Version 4.0 (Carl Zeiss Meditec, Inc.). The area under the receiver operating characteristic (AROC) was used to compare the glaucoma diagnostic performance of anatomic parameters and their combinations. Or-logic, support vector machine (SVM), relevance vector machine (RVM), and linear discrimination function (LDF) were used to combine diagnostic parameters. Eighteen-fold cross-validation was used to train and test the combination methods.
The top 3 cpNFL thickness parameters are overall, inferior quadrant and superior quadrant with AROC values of 0.89, 0.882 and 0.857, respectively. The top 3 ONH parameters were horizontal integrated rim width (HIRW), vertical integrted rim area (VIRW), and cup/disc vertical ratio (CDVR), with AROC values of 0.904, 0.90 and 0.883, respectively. All MT parameters had AROC values of 0.81 or less. Or-logic combination of the top 3 cpNFL parameters provided an AROC of 0.922. We chose to combine the top 3 cpNFL parameters, along with HIRW, VIRW, and CDVR to optimize glaucoma diagnosis. SVM gives the highest AROC value of 0.939, followed by or-logic (0.935), LDF (0.927) and RVM (0.924). SVM is significantly better than RVM (p=0.01) and LDF (p=0.02) although it may not be clinically significant.
cpNFL and ONH parameters have much better diagnostic accuaracy than MT measurements. Combining cpNFL and ONH parameters provided the best diagnostic performance.
This PDF is available to Subscribers Only