Progression of Patterns (POP) Using Independent Component Analysis Identifies Glaucoma Progression

Abstract

 

Purpose:

 

We evaluated Progression of Patterns (POP) for its ability to identify and quantify progression of glaucomatous visual field defects.

 

Methods:

 

POP uses the variational Bayesian independent component mixture model (VIM) [TAOS 2009]. VIM separated Swedish Interactive Thresholding Algorithm (SITA) visual fields (VFs) from a set of 2085 normal and glaucoma eyes into 9 axes of maximally different patterns of VF defects. POP used a second set of 55 eyes with 5 VFs collected within 4 weeks (D.R. Anderson) to simulate stability and a third set of 4,186 VFs from 628 eyes from the Diagnostic Innovations in Glaucoma Study and the African Descent and Glaucoma Evaluation Study to test for progression (mean+sd 6.67+1.66 VFs followed 4.01+1.40 years). Eyes were classified into diagnostic categories based on VFs and optic disk photographs and whether they had photographic evidence of progressive glaucomatous optic neuropathy (PGON). Each sequence of fields was projected on 7 VIM glaucoma axes. POP concentrated on change detection of the VF defect axis with maximal change. Using linear regression (LR), which incorporated 95% confidence limits of stable eyes and variability in an evaluated eye’s sequence of VFs, the probability of progression in the third set of eyes was estimated for POP and compared to the LR of Humphrey Visual Field Index (VFI), and AGIS scores.

 

Results:

 

The evaluated populations had suspicious disk or pressure (suspects), glaucomatous fields (VF only), glaucomatous fields and GON (VF+GON), or PGON. Table shows percent progressing at specificity set at 95%.

 

Conclusions:

 

The automatic algorithm in POP 1) could give probability of progression, 2) identified more progressing eyes in groups at higher risk for progression, and 3) found change of a specific VIM defect pattern to be a sensitive indicator of progression in more progressing eyes than did change in global indices.  

 

Keywords: visual fields • clinical (human) or epidemiologic studies: prevalence/incidence