Pattern Recognition Reveals Unique Sensitivity Patterns Useful for a Hemifield Test Analysis in Glaucoma

Michael Kalloniatis

Abstract

Purpose: Cluster analysis (pattern recognition) reveals unique areas of contrast sensitivity within the visual field (VF). We compared how well the groups of points used in the Glaucoma Hemifield Test (GHT) derived from nerve fiber layer (NFL) distributions detect glaucomatous VF defects (VFDs) compared to those derived from contrast sensitivity isocontours (CSIs) found using pattern recognition.

Methods: The Humphrey Field Analyzer (HFA) was used in full threshold mode to measure thresholds across the 30-2 test grid using the Goldmann size III target in 60 normal subjects. Thresholds were converted to a 50 year old equivalent observer to generate multiple group-averaged data sets.  Data sets were converted to pixel values and analyzed using unsupervised classification (ISODATA or k-means clustering: PCI, Canada).  We derived CSIs (delineating points with the same contrast sensitivity) separated with an accuracy of >96%. We then compared clusters of points using the conventional GHT with those derived from the CSIs in 50 VFs of 27 early-to-moderate (mean deviation (MD) > -6 dB) glaucoma patients, and present a model for performing asymmetry analysis to detect VFDs.

Results: The clusters of points used by the GHT derived from NFL distribution patterns and those derived from CSIs are discordant with few areas showing similar distribution patterns. This is expected given the different underlying methods used to derive the clusters. Asymmetry analysis using grouped points with the same CSI within each hemifield detected more VFDs (64%) compared to that reported using the HFA GHT (38% of the glaucoma cohort). A subset of 5 VFs from 5 eyes with moderate glaucoma (-6 dB ≤ MD < -12 dB) confirmed that the model performed just as well with more severe VFDs (100%).

Conclusion(s): CSIs derived using pattern recognition analysis provides a framework for new methods of grouping test locations to improve detection of glaucomatous VFDs in early stages of the disease. 

Details

Year: 2017

Program Number: 170040

Resource Type: Scientific Program

Author Affiliation: Centre for Eye Health and SOVS, UNSW Australia

Co-Authors: Jack Phu, Sieu Khuu

Co-Author Affiliation: Centre for Eye Health SOVS UNSW, SOVS UNSW

Room: E354A