Purchase this article with an account.
Siamak Yousefi, Akram Belghith, Michael Goldbaum, Linda Zangwill, Felipe Medeiros, Robert Weinreb, Renato Lisboa, Christopher Bowd, Hamilton Glaucoma Center; Quadratic Bayesian Pattern Detection for Detecting Glaucomatous Change in Follow-up SD-OCT RNFL Thickness Measurements. Invest. Ophthalmol. Vis. Sci. 2013;54(15):4843. doi: https://doi.org/.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
We proposed a quadratic Bayesian pattern detector to detect change from longitudinal series of SD-OCT measurements, including retinal nerve fiber layer (RNFL) thickness measured in 6 sectors and average RNFL thickness (a total of 7 input dimensions), in stable and progressing glaucoma eyes.
A quadratic Bayesian pattern detection method was employed to detect change in the seven-dimensional vector of SD-OCT based-RNFL thickness measurements, by comparing baseline to follow-up exams, from a longitudinal series. Spectralis RNFL image series were obtained from non-progressing glaucoma eyes and progressing glaucoma eyes which correspond to no-change and change, respectively. The dataset included 331 images from 20 non-progressing glaucoma patients (imaged once a week for 5 consecutive weeks) and 81 images from 20 progressing glaucoma patients from the UCSD Diagnostic Innovations in Glaucoma Study (DIGS). Progression was defined by standardized assessment of stereophotographs and/or by designation as “likely progression” based on SAP Guided Progression Analysis (mean follow-up 4 years, 4 tests). Two normative databases were generated from 90% of the data in the original datasets by calculating the norm 1 distance between the RNFL measurements from the baseline and each follow-up exam, for each group. The quadratic Bayesian pattern detector was trained based on the normative datasets and then was tested using ten-fold cross-validation to evaluate the accuracy of change detection. For testing, for each subject, each follow-up exam was compared to the baseline exam separately. If more than half of the follow-up exams showed progression based on the trained Bayesian pattern detector, the eye was classified as progressed, otherwise, the eye was classified as non-progressed. Sensitivity and specificity were computed based on this rule.
The diagnostic accuracies were: 80% sensitivity in progressors and 78% specificity in non-progressors.
A Bayesian pattern detection method was developed that could detect glaucomatous progression from baseline and follow-up images. This method provided reasonable specificity to detect stable eyes and sensitivity to detect progressing eyes, with the potential to improve accuracy by increasing number of samples in the normative database.
This PDF is available to Subscribers Only