Mixture of factor analysis of standard visual fields

MIXTURE OF FACTOR ANALYSIS OF STANDARD VISUAL FIELDS

PAMELA A. SAMPLE,1 KWOKLEUNG CHANG,3 CATHERINE BODEN,1 TE-WON LEE,3 ROBERT N. WEINREB,1 TERRENCE SEJNOWSKI2 and MICHAEL H. GOLDBAUM1

1Glaucoma Center, Department of Ophthalmology, University of California at San Diego; 2Computational Neurobiology Laboratory, Salk Institute; 3Institute for Neural Computation, University of California at San Diego; La Jolla, CA, USA

Abstract

Purpose

To extract patterns of field loss using variational bayesian mixture of factor analysis (vbMFA).

Methods

Standard perimetry absolute threshold values for 52 locations plus age from one eye each of 156 patients diagnosed with glaucomatous optic neuropathy (GON) and 189 normals were evaluated with unsupervised vbMFA to separate the fields into clusters. Fields were not used to select subjects.

Results

The vbMFA formed four distinct clusters. The ‘normal cluster’ held 186 normals + 45 patients. Each GON cluster could be represented by a ‘typical’ pattern of defect: GON1 (56 patients + 3 normals) by a general reduction with MD of –2.65 ± 1.72; GON 2 (39 patients) by a superior hemifield defect, and GON 3 (16 patients) by an inferior hemifield defect with or without superior field involvement. Specificity was 98%; sensitivity was 71%.

Conclusions

vbMFA accurately clustered patients into groups with typical glaucomatous patterns of loss. vbMFA may be very helpful for learning patterns of defect in newer psychophysical tests that have different numbers and locations for test stimuli.

Acknowledgment

Supported by NIH Grants EY08208 (PAS), EY13235 (MHG), Research to Prevent Blindness (PAS), Howard Hughes Medical Institute (TS). Commercial relationships: none.

Address for correspondence: Pamela A. Sample, PhD, Department of Ophthalmology, University of California, San Diego, 9500 Gilman Drive, La Jolla, CA 92093-0946, USA. Email: psample@eyecenter. ucsd.edu

Perimetry Update 2002/2003, p. 35

Proceedings of the XVth International Perimetric Society Meeting, Stratford-upon-Avon, England, June 26–29, 2002

edited by David B. Henson and Michael Wall

© 2004 Kugler Publications, The Hague, The Netherlands

VARIABILITY COMPONENTS OF STANDARD PERIMETRY

PAUL G.D. SPRY1,2 and CHRIS A. JOHNSON1

1Discoveries in Sight, Portland, OR, USA; 2Bristol Eye Hospital, Bristol, UK

Abstract

Background

Responsiveness to light stimuli at a single test location varies according to a psychometric function, the frequency-of-seeing curve. This function can be used to precisely define threshold (50% detection level) and response variability (slope). Although the relationship between threshold sensitivity and response variability in glaucoma has been described by a number of investigators, 1-3 the temporal behavior of other frequency-of-seeing curve parameters, or variability components, remain uncharacterized. Specifically, it is possible that the frequency-of-seeing curve gradient from a particular test location may not remain constant and could change in shape over relatively short time periods. Furthermore, it is also possible that the curve may retain a constant shape but may shift laterally in sensitivity space.

Purpose

To examine the temporal behavior of precisely measured threshold and response variability.

Methods

Eight trained normal individuals and seven glaucoma patients were examined using standard perimetry. The method of constant stimuli was used to test three test locations in a single eye of each subject using an externally controlled Humphrey Field Analyzer I. Subjects were tested once weekly at the same test locations for five consecutive weeks. Frequency-of-seeing curves were constructed by fitting data with cumulative gaussian functions and were used to define threshold (50% detection level, dB) and response variability (frequency-of-seeing curve inter-quartile range, dB). The distribution of both of these parameters (inter-quartile range, dB) over the study period was used to quantify their temporal behavior.

Results

As reported by previous investigators, response variability was found to be higher (shallower frequency- of-seeing curve slope) in glaucoma patients than normal individuals (p < 0.001). Additionally, variations in both threshold sensitivity and response variability were observed and found to be significantly greater in glaucoma patients than normal individuals.

In glaucoma patients, mean (standard deviation) response variability was 6.2 (5.0) dB and was greater than both mean threshold sensitivity variation and mean frequency-of-seeing curve slope variation, which were measured as 2.5 (2.1) dB and 3.3 (3.4) dB, respectively. No significant difference existed between the amount of threshold sensitivity variation and frequency-of-seeing curve slope variation, although both were inversely related to mean threshold sensitivity.

Address for correspondence: Paul G.D. Spry, PhD, Bristol Eye Hospital, Lower Maudlin Street, Bristol, BS1 2LX, UK. Email: paul.spry@ubht.swest.nhs.uk

Perimetry Update 2002/2003, pp. 37–38

Proceedings of the XVth International Perimetric Society Meeting, Stratford-upon-Avon, England, June 26–29, 2002

edited by David B. Henson and Michael Wall

© 2004 Kugler Publications, The Hague, The Netherlands

Changes in threshold sensitivity and response variability for test locations tested within each individual did not occur in parallel for the majority of subjects.

Discussion

In this experiment, three components of threshold measurement variability have been described. The first of these is the slope of the psychometric function (frequency-of-seeing curve). Data obtained in this experiment for this variability component, response variability, are in agreement with that reported by previous investigators, who have shown that the frequency-of-seeing curve becomes shallower with increases in threshold (reductions in sensitivity) that occur in glaucoma.1-3 The other two variability components we have described occur over a longer time period than response variability. These longerterm components consist of change in threshold and change in response variability over time. Both these longer-term variability components appear to increase with threshold in patients with glaucoma, although to a significantly lesser extent than shallowing of the curve slope. No evidence was found to suggest that changes in slope or sensitivity at one test location in any given field were similar to changes occurring at other test locations elsewhere within the same visual field.

References

1.Weber J, Rau S: The properties of perimetric thresholds in normal and glaucomatous eyes. German J Ophthalmol 1: 79-85, 1992

2.Chauhan BC, Tompkins JD, LeBlanc RP, McCormick TA: Characteristics of frequency-of-seeing curves in normal subjects, patients with suspected glaucoma, and patients with glaucoma. Invest Ophthalmol Vis Sci 34:3534-3540, 1993

3.Henson DB, Chaudry S, Artes PH, Faragher EB, Ansons A: Response variability in the visual field: comparison of optic neuritis, glaucoma, ocular hypertension, and normal eyes. Invest Ophthalmol Vis Sci 41:417-421, 2000