Purchase this article with an account.
Juan Xu, Hiroshi Ishikawa, Gadi Wollstein, Richard A. Bilonick, Larry Kagemann, Ian A. Sigal, Jonathan Grimm, James G. Fujimoto, Joel S. Schuman; Glaucoma Detection Using Super Pixel Analysis of Macular Three-Dimensional (3D) Spectral Domain Optical Coherence Tomography (SD-OCT). Invest. Ophthalmol. Vis. Sci. 2012;53(14):4109.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To assess glaucoma discrimination of a novel super pixel machine classifier analysis technique using macular 3D SD-OCT data.
125 eyes of 66 subjects (28 healthy, 45 glaucoma suspect and 52 glaucomatous eyes) were scanned with SD-OCT (Cirrus HD-OCT; Carl Zeiss Meditec, Inc., Dublin, CA; Macular 200×200 scan). Self-size-adjusting super pixels were automatically segmented on 2D macular retinal ganglion cell layer-inner plexiform layer (GCIPL) thickness maps by grouping homogeneous neighbor pixels (in terms of GCIPL thickness) using our custom software (Figure). 67 super pixel features, such as average intensity value and size, were extracted and ranked with a forward feature selection algorithm in terms of maximum relevance and minimum redundancy. Feature selection was used for ranking the parameters and identifying useful inputs with 2 machine classifiers: boosting learning and support vector machine (SVM). 10-fold cross validation area under the receiver operating characteristics (AUC) of the machine classifier outputs were compared with the average GCIPL thickness and corresponding peripapillary retinal nerve fiber layer (RNFL) super pixel analysis for discriminating between the diagnostic groups.
Discrimination between healthy vs. glaucoma suspects with GCIPL thickness was statistically significantly improved with the boosting method (AUC 0.696 to 0.926, p<0.001, DeLong test; Table). No statistically significant difference was detected comparing the best AUCs of macular analysis with peripapillary RNFL super pixel analysis.
A new macular 3D OCT analysis technique improved the discrimination between healthy and glaucoma suspect eyes over the traditional GCIPL analysis and performed as well as peripapillary RNFL super pixel analysis. This new method has the potential to improve early detection of glaucomatous damage.
This PDF is available to Subscribers Only