Purchase this article with an account.
D. B. Henson; Development of a New Algorithm for Computerised Threshold Perimetry Using Computer Simulation. Invest. Ophthalmol. Vis. Sci. 2010;51(13):5493.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To report on the performance of a new algorithm for threshold perimetry that uses relaxed terminating criteria (TC) at locations where sensitivity is normal or severely depressed and which starts from prior threshold estimates.
Computer simulation was used to determine the number of presentations and the distribution of errors using an adaptive Bayesian algorithm (ZEST) to establish threshold estimates, similar to SITA. A 24-2 visual field test was simulated with either a constant TC ( standard deviation of the probability density function) or TC that were relaxed at locations where the sensitivity is within the normal range (>normal -4dB) or where it is severely depressed (<10dB). Starting levels were either normal age matched sensitivity values or values obtained from the patients prior test. The simulation used threshold and error rates from a database of 179 normal and 93 glaucomatous eyes (mean MD -4.9, range -21.5 to 1; mean PSD 5.1, range 1.6 to 11.9). The fixed/non-relaxed TC was selected to give an equivalent number of presentations to those in SITA standard and SITA fast on normal eyes when using a constant TC and normal starting values.
The table gives the number of presentations for each algorithm. Using prior threshold estimates in glaucomatous eyes not only reduced the number of presentations but also the bias (mean error) from 1.12 to 0.34dB and 1.29 to 0.38dB for the Standard and Fast TC and reduced the 95% confidence limits of the errors from 6.11 to 5.45dB and 6.78 to 6.11dB. Using a relaxed TC of 5.5 at normal and severely damaged locations, where precision is less important, further reduced the number of presentations by up to 64%.
Test time can be reduced in eyes with glaucoma by starting the test from prior threshold estimates. Using prior data also improves the accuracy (bias and error distribution) of the threshold estimates. Using relaxed TC at locations where precision is less important further reduces the number of presentations and overall test time by as much as 64% with no loss in precision at locations where there is relative sensitivity loss.
This PDF is available to Subscribers Only