A Novel Simulation Tool for the Evaluation of Perimetric Algorithms: Benefits of Using Prior Data at Test Onset

Abstract

 

Purpose

 

To describe a simulation tool for evaluating the performance of perimetric algorithms and to use this tool to report on the benefits of using prior data at the onset of a test with Full Threshold (FT) and Zippy Adaptive Threshold Algorithm (ZATA) algorithms.

 

Methods

 

Visual field data from 15 stable glaucoma patients, 3 for each one of Brusini’s glaucomatous stages (1 to 5), were collected from a previously extracted database. The mean threshold of the last 5 tests was used to determine each patient’s “true” threshold at all 24-2 test locations. The simulation tool uses the patient response characteristics defined by Henson et al (Invest Ophthal Vis Sci 2000;41:417-421) and sets error rates for each trial to a randomly selected case from a database of 150 eyes. Two different algorithms, FT and ZATA, were tested starting either from age-corrected normative values or a prior set of data. Each of the 15 input visual fields was simulated 20 times. For the analysis of data, the two locations adjacent to the blind spot were excluded. The distribution of error, that is the subtraction of the measured sensitivity from the “true” one, and the number of stimulus presentations were recorded.

 

Results

 

Each simulated test took approximately 10s to complete. The distribution of errors for FT across the range of “true” thresholds (0-31 dB) in 5dB increments showed that the use of a prior set of data results in better accuracy, reducing the median error in all increments apart from the extreme ones (0-5 dB and 26-31 dB); see figure 1. The use of prior data significantly reduced the number of presentations for stages 3-5; see figure 2. ZATA required fewer presentations than FT for almost all defect levels but most notably for those in stages 1-2.

 

Conclusions

 

Visual field simulations with a pre-defined set of defects can be used to rapidly assess and compare the performance of threshold algorithms and the benefits or otherwise of adaptations such as the use of prior data. They overcome many of the shortcomings of patient based trials such as the lack of information on the true extent of loss and sampling bias.

 

 

Fig 1:Box is 25-75%, band is median, whiskers 1.5 IQR.* symbol shows the threshold increments where significant (p<0.01) reduction of the median error due to prior data was observed.

 

 

Fig 2:* symbol shows the glaucomatous stages that significant reduction of stimulus presentations was observed.

 

Keywords: 758 visual fields • 473 computational modeling • 468 clinical research methodology