May 2008
Volume 49, Issue 13
ARVO Annual Meeting Abstract  |   May 2008
Stochastic Optimization Framework for the Optimization of Prosthetic Vision
Author Affiliations & Notes
  • W. Fink
    Doheny Eye Institute, University of Southern California, Los Angeles, California
  • M. A. Tarbell
    Visual and Autonomous Exploration Systems Research Laboratory, California Institute of Technology, Pasadena, California
  • Footnotes
    Commercial Relationships  W. Fink, Caltech, P; M.A. Tarbell, Caltech, P.
  • Footnotes
    Support  DOE Award DE-FG02-06ER64310
Investigative Ophthalmology & Visual Science May 2008, Vol.49, 1779. doi:
  • Views
  • Share
  • Tools
    • Alerts
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      W. Fink, M. A. Tarbell; Stochastic Optimization Framework for the Optimization of Prosthetic Vision. Invest. Ophthalmol. Vis. Sci. 2008;49(13):1779. doi:

      Download citation file:

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

  • Supplements

Purpose: : To introduce a Stochastic Optimization Framework (SOF) for (blind) subject-in-the-loop optimization of (electric) stimulation patterns for visual prostheses.

Methods: : Akin to the convergent process of obtaining an optimal and individually tailored prescription for eyeglasses, subjects are presented with stimulation patterns, mapped to a pixel array of a certain size (e.g., 4 x 4, 8 x 8, or higher), where each pixel of the array represents, for example, an electrode of a retinal implant. The set of stimulation parameter values (e.g., pulse amplitude, pulse shape, pulse onset, pulse duration, etc.) that characterize the behavior of each electrode is represented for the purpose of this study as a one-dimensional variable with 256 discrete values, i.e., grayscale values. In an iterative, three-alternate forced-choice test set up, subjects repeatedly have to decide and choose whether a Current Optimal Perception, or a New Alternate Perception, both generated by respective sets of stimulation parameter values provided by a multivariate optimization algorithm, or neither perception, is closer to a given Target Perception. This process continues until a useful approximation of the Target Perception is achieved.

Results: : We have developed a software platform with Graphical User Interface that permits testing, evaluating, and benchmarking of (blind) subject-in-the-loop multivariate optimization algorithms to define and optimize (electric) stimulation patterns for visual prostheses to best represent an object or scene to the individual (blind) subject. Tests conducted with a Simulated-Annealing-related multivariate optimization algorithm have not only demonstrated feasibility of the SOF approach but have also shown robustness against erroneous subject feedback.

Conclusions: : The Stochastic Optimization Framework, introduced here for the optimization of stimulation patterns for retinal implant electrode arrays, can be readily expanded to higher-dimensional parameter spaces, e.g., larger numbers of electrodes with more individual parameters. Once several objects/scenes have been successfully presented to subjects via the SOF, artificial neural networks can be employed to extract and learn signature stimulation patterns that would allow for a useful presentation of previously unseen objects/scenes to the subjects via these learned signature stimulation patterns. This may provide insight in the residual retinal processing capabilities of electric stimulation through visual prostheses.

Keywords: perception • image processing • retinal connections, networks, circuitry 

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.