A Perimetric Re-Test Algorithm That is Significantly More Accurate Than Current Procedures


Andrew Turpin, School of Computer Science and Information Technology, RMIT University, Melbourne, Australia

Allison McKendrick and Darko Jankovic, Department of Optometry and Vision Science, University of Melbourne, Melbourne, Australia


Status

To be presented at XVIIth International Visual Field & Imaging Symposium, Portland, Oregon, USA July 11-14, 2006.

This work builds on our OVS paper which reports on simulation work with a test procedure that combines supra-threshold and full-threshold procedures to tradeoff test time and accuracy. Using a similar idea, this abstract reports on simulations for a re-test algorithm that combines supra-threshold and full-thresholding procedures.


Abstract

Purpose: To develop a fast and accurate perimetric test procedure that makes use of threshold information derived from previous tests.

Methods: Computer simulation using 4 levels of patient error (none, false-positive, false-negative, and unreliable) was used to analyse several retest algorithms. Baseline systems included simply running two existing threshold algorithms that do not make use of previous information, Full Threshold (FT) and ZEST (Z); allowing ZEST to continue from the previous test without re-initialising the input pdf (Z-CONT); and running ZEST with a Gaussian pdf about the previous measured threshold (Z-GAUSS). Our new algorithm (REMU) is based on our previously published EMU algorithm [1] which combines supra-threshold and ZEST threshold procedures to trade off test time and test accuracy. Specifically, a supra-threshold test is made for each location based on the previous threshold at that location with a general-height correction. If two or two of three supra-threshold presentations are seen, then the threshold reported is same as previous, otherwise a ZEST is performed using a Gaussian pdf centred 2 dB below the failed supra-threshold value. The pdf has an initial standard deviation of 3 dB, and the ZEST is terminated when the standard deviation falls below 1.5 dB.

Results: When there is no change in field from test to retest, the retest algorithms (Z-GAUSS, Z-CONT and REMU) are all at least one presentation per location faster than the test algorithms, on average, and at least 0.5 dB more accurate. When fields change from test to re-test, REMU is faster and more accurate than the other two retest approaches.

Conclusions: The new retest algorithms are as fast and more accurate than current gold-standard test algorithms such as SITA, and REMU is more accurate than Z-CONT and Z-GAUSS when fields change from one visit to the next.