May 2003
Volume 44, Issue 13
ARVO Annual Meeting Abstract  |   May 2003
EMU: A New Algorithm for Automated Perimetry
Author Affiliations & Notes
  • A.H. Turpin
    School of Computing, Curtin University, Perth, Australia
  • A.M. McKendrick
    School of Psychology, University of Western Australia, Perth, Australia
  • Footnotes
    Commercial Relationships  A.H. Turpin, None; A.M. McKendrick, None.
  • Footnotes
    Support  NHMRC Australian Clinical Research Fellowship #139150 (AMM)
Investigative Ophthalmology & Visual Science May 2003, Vol.44, 70. doi:
  • Views
  • Share
  • Tools
    • Alerts
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      A.H. Turpin, A.M. McKendrick; EMU: A New Algorithm for Automated Perimetry . Invest. Ophthalmol. Vis. Sci. 2003;44(13):70.

      Download citation file:

      © ARVO (1962-2015); The Authors (2016-present)

  • Supplements

Abstract: : Purpose: In general it is not possible to measure accurate estimates of sensitivity at 54 locations in a visual field using 3 or 4 presentations per location. Current algorithms such as Full Threshold (FT) and SITA sacrifice accuracy at non-normal threshold values in order to achieve acceptable test times. For detecting glaucomatous visual field loss and its progression it would be preferable to sacrifice the accuracy of normal threshold values, maintaining accuracy for glaucomatous points. We present an algorithm that achieves this aim. Methods: Our algorithm uses a supra-threshold test to locate non-normal locations, and then accurately thresholds these locations. We tested, using computer simulation, many supra-threshold and thresholding schemes in combination, and report the best as algorithm EMU. EMU shows 2 supra-threshold stimuli at each location. If either of the stimuli are missed, then a ZEST procedure with a uniform pdf is performed at that location, terminating when the pdf has a standard deviation (SD) of 1.5 dB or less. The test procedures were evaluated using a 24-2 test pattern. Input to the simulation were visual fields from 265 normal patients and 163 glaucomatous patients using 3 error models: ideal, typical and unreliable. Typical patients had a 15% false positive and negative response rate and a random variation of threshold according to a Gaussian distribution with SD of 1 dB. Results: Mean No. presentations on typical normal patients: EMU 4.21, FT 5.06. Mean No. presentations on typical glaucoma patients: EMU 5.51, FT 5.71. For points in a typical glaucomatous visual field that had true sensitivities below the 95% lower limit for normal (corrected Hill of Vision): EMU had a mean error of -0.2dB, SD = 3.0dB; FT had a mean error of -0.2dB, SD = 4.2dB. The difference in variability (SD of error) is much more pronounced for low threshold values. For example, for true thresholds of 8dB in typical glaucoma patients, mean error was: EMU -0.6dB (SD = 2.2dB); FT -1.8dB (SD = 5.4dB). Conclusions: EMU is faster than FT, and much less variable than FT is areas of low sensitivity. EMU may provide sensitivity estimates that are better at tracking glaucomatous progression than current techniques.

Keywords: perimetry • visual fields • computational modeling 

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.