Gaussian Mixture Model Expectation Maximization for Glaucoma Progression Detection: An open-source R Package

Abstract

Purpose : To develop and assess the performance of an open-source R package for glaucomatous visual field change detection and monitoring progression (VFPro)

Methods : VFPro was trained on archival cross-sectional visual fields to identify 20 prevalent patterns of visual field defects in glaucoma using Gaussian Mixture Model Expectation Maximization and Principal Component Analysis. VFPro detects progression along these glaucomatous patterns of visual field defect. VFPro parameters were derived from a test-retest dataset including 133 eyes tested 10 times over 10 consecutive weeks to mimic no glaucomatous change. VFPro was then integrated into an R package (R programming language) and was made publicly available on the web. VFPro R package was applied on 270 VF sequences from 270 glaucomatous eyes from Rotterdam Eye Hospital in Netherland and was compared against global and pointwise analyses (parameters derived from the same test-retest dataset at 95% specificity). Survival analysis was used to compare all methods.

Results : At equal 95% specificity, VFPro identified more eyes as progressed followed by pointwise and global analyses. VFPro detected eyes earlier than pointwise and global, particularly towards the end of the survival curve, indicating identification of more slowly progressing eyes earlier than pointwise and global analyses.

Conclusions : VFPro software could be downloaded and used freely for glaucoma progression detection in glaucoma clinical care and research.

This is an abstract that was submitted for the 2018 ARVO Annual Meeting, held in Honolulu, Hawaii, April 29 - May 3, 2018.

 

Figure 1. Time to detect progression using global, pointwise, and VFPro analyses in a dataset including 270 glaucoma eyes. VFPro detected more eyes as progressed, particularly, more slowly progressing eyes than other methods.