Purchase this article with an account.
Akram Belghith, Christopher Bowd, Robert Weinreb, Andrew Tatham, Atsuya Miki, Felipe Medeiros, Linda Zangwill; Glaucoma progression detection using a Bayesian-fuzzy logic approach applied to 3D spectral domain optical coherence tomography optic nerve head images. Invest. Ophthalmol. Vis. Sci. 2013;54(15):4798. doi: https://doi.org/.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To detect glaucomatous changes in the optic nerve head (ONH), we proposed a novel fuzzy Bayesian detection scheme (FBDS), which aims to classify the 3D Spectralis SD-OCT images of the ONH into "non-progressing" and the "progressing" glaucoma classes.
We formulated the detection of glaucomatous change between baseline image and follow up images as a missing data problem. We proposed the use of a Bayesian approach to model the a priori we have on the change detection map. We used the Markov Random Field Model (MRF) to handle the spatial dependency of changed pixels. To accommodate the presence of false positive detection, the estimated change detection map is then used to classify a 3D spectralis SDOCT image into the "non-progressing" and "progressing" glaucoma classes based on a novel fuzzy logic classifier. An independent training set, which consists of 10 normal eyes, 5 non-progressing eyes and 10 progressing eyes, was used to train the fuzzy logic classifier. Progression was defined as the presence of one or more changed follow-ups. Diagnostic accuracy was estimated using 117 eyes of 75 participants from the UCSD Diagnostic Innovations in Glaucoma Study (DIGS). Sensitivity was estimated in 27 eyes classified as progressing by standardized assessment of stereophotographs by 2 independent graders and/or by designation as “likely progression” based on visual field Guided Progression Analysis (mean follow-up 4 years, 4 tests). Specificity was estimated using 50 stable glaucoma eyes (imaged once a week for 5 consecutive weeks) and using 40 healthy eyes (mean follow-up 3 years, 3 tests). We compared the FBDS method to a Support vector Machine SVM with the whole image difference between the baseline and a follow up as input, and to 2 alternative FDBS methods, one without the MRF prior and one with a threshold classifier.
Sensitivity in progressors, specificity in stable Glaucoma eyes and specificity in normals, obtained with different methods, are presented in Table1.
The FBSD method with use of the MRF as prior and the fuzzy classification results in high specificity in both normal and stable glaucoma eyes (94% and 92% respectively) while maintaining good sensitivity.
This PDF is available to Subscribers Only