June 2013
Volume 54, Issue 15
Free
ARVO Annual Meeting Abstract  |   June 2013
Automated stimulus choice in condensed grids for assessment of visual field defects
Author Affiliations & Notes
  • Luke Chong
    Department of Optometry and Vision Sciences, The University of Melbourne, Parkville, VIC, Australia
  • Allison McKendrick
    Department of Optometry and Vision Sciences, The University of Melbourne, Parkville, VIC, Australia
  • Andrew Turpin
    Department of Computing and Information Systems, The University of Melbourne, Parkville, VIC, Australia
  • Footnotes
    Commercial Relationships Luke Chong, None; Allison McKendrick, Heidelberg Engineering GmbH (F); Andrew Turpin, Heidelberg Engineering (F)
  • Footnotes
    Support None
Investigative Ophthalmology & Visual Science June 2013, Vol.54, 3943. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Luke Chong, Allison McKendrick, Andrew Turpin; Automated stimulus choice in condensed grids for assessment of visual field defects. Invest. Ophthalmol. Vis. Sci. 2013;54(15):3943.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose: A condensed stimulus grid improves spatial characterisation of visual field (VF) defects (Schiefer et al IOVS 2010). We define a new algorithm that autonomously selects locations for stimulus presentations within a dense grid to improve characterisation of VF defects. We hypothesised our algorithm would provide more accurate estimates of threshold for all locations than linear interpolation of a less dense grid.

Methods: Each location in our grid runs a ZEST-like procedure, with seed locations initially presented twice, and subsequent trials taking place at the mid-point of the two locations with the highest gradient (estimated threshold/distance). After each presentation, locations that have had no presentations are updated based on a weighting of spatial relationships determined from a database of 297 VFs. The total number of presentations was limited to be comparable to existing procedures. The algorithm was tested using computer simulation on 54 empirical VFs (HFA 24-2). A 5x5 (25 locations) grid of 24-2 points was selected from each field, from which every second location was omitted to create a 3x3 (9 locations) grid. Omitting locations models a coarser VF grid where the intermediate values are known. Three results were compared: 1) Our algorithm run on all 25 locations; 2) our procedure for the 3x3 grid with untested locations being "interpolated estimates"; 3) the input thresholds as estimates on the 3x3 grid with untested locations being "interpolated from true-fields". Procedures were compared to the input VF to determine error. Simulations were repeated 200 times with stochastic patient responses. Locations derived from interpolation were compared to corresponding measured locations in the 5x5 grid for analysis.

Results: The median absolute error (MAE) was lower in the 5x5 grid compared to the interpolated 3x3 grid (2.57dB vs 3.15dB, Wilcoxon P < 0.05). The interpolated true field had lower MAE than the 5x5 grid (1.50dB vs 2.57dB, Wilcoxon P < 0.05). The interquartile ranges for the 5x5 grid, “interpolated estimates” and “interpolated from true-fields” were 1.76dB, 3.52dB and 2.50dB respectively.

Conclusions: This procedure shows potential to automatically choose additional locations to test within areas of loss without increasing test time. Simulations suggest testing locations spaced closer together provides more accurate information about a scotoma than interpolating across a coarser grid.

Keywords: 642 perimetry • 473 computational modeling • 758 visual fields  
×
×

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.

×