Purchase this article with an account.
Zhong-Lin Lu, Fang Hou, Luis A Lesmes, Woojae Kim, Hairong Gu, Mark Pitt, Jay Myung; A large-sample study for evaluating the precision of the quick CSF method. Invest. Ophthalmol. Vis. Sci. 2015;56(7 ):3899.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
The quick CSF method (Lesmes, et al, 2010) applies a Bayesian adaptive algorithm to estimate the contrast sensitivity function (CSF) with high precision and reduced testing time (~5 min). We collected a large dataset of CSF to 1) determine reliability as a function of test duration, 2) evaluate the concordance between estimates against intra- and inter-run variability (via Bayesian confidence and repeated testing, respectively), and 3) conduct a power analysis for detecting CSF change.
CSFs of 112 college students with normal vision were repeatedly assessed using quick CSF with a 10-letter identification task. For each observer, running CSF estimates were calculated for each trial, via bootstrap statistics for the area under the log CSF (AULCSF), computed by resampling from the Bayesian posterior distribution of the CSF.
1) After 6 trials, the AULCSFs from the two repeated measurements were significantly correlated. Pearson’s r increased from 0.22 (p=0.02) to 0.84 (p<0.001) as trial number increased from 6 to 50. The 95% confidence interval of the ratio between the two AULCSFs was [0.94, 1.02] at trial 6 and [0.99, 1] at trial 50.<br /> 2) The comparable metrics of intra- and inter-run variability provided by standard deviations of AULCSF estimates were 0.20 and 0.25 log units after 10 trials, 0.13 and 0.16 log units after 20 trials, and 0.07 and 0.09 log units after 50 trials, respectively.<br /> 3) From the posterior distributions of the CSFs, we computed the minimum AULCSF difference (MAD) that can be detected by quick CSF with 95% posterior probability as a function of both trial and observer numbers (Figure 1). To detect MADs of 0.2, 0.1 and 0.05 l log units with 25 quick CSF trials, we needed to run 2, 6 and 27 observers, respectively. To detect the same MADs in 50 trials, only 1, 3 and 11 observers were needed. With 20 observers, we needed 5, 11, and 30 trials to detect MADs of 0.2, 0.1 and 0.05 log units, respectively. With 112 observers, we needed only 3, 6 and 12 trials to detect the same MADs, respectively.
The quick CSF method is very precise and highly reliable. The high precision and reliability make it possible to use the method to efficiently measure CSF and detect CSF changes with greatly reduced sample size and costs in clinical trials.
This PDF is available to Subscribers Only