June 2021
Volume 62, Issue 8
Open Access
ARVO Annual Meeting Abstract  |   June 2021
An explainable AI model to predict electrode deactivation in retinal prostheses from routine clinical measures
Author Affiliations & Notes
  • Michael Beyeler
    Psychological & Brain Sciences, University of California Santa Barbara, Santa Barbara, California, United States
    Computer Science, University of California Santa Barbara, Santa Barbara, California, United States
  • Zuying Hu
    Computer Science, University of California Santa Barbara, Santa Barbara, California, United States
  • Footnotes
    Commercial Relationships   Michael Beyeler, None; Zuying Hu, None
  • Footnotes
    Support  None
Investigative Ophthalmology & Visual Science June 2021, Vol.62, 3212. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Michael Beyeler, Zuying Hu; An explainable AI model to predict electrode deactivation in retinal prostheses from routine clinical measures. Invest. Ophthalmol. Vis. Sci. 2021;62(8):3212.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : To provide appropriate levels of stimulation, retinal prostheses must be calibrated to an individual's perceptual thresholds ("system fitting"). Nonfunctional electrodes may then be deactivated to reduce power consumption and improve visual outcomes. However, thresholds vary drastically not just across electrodes but also over time, thus calling for a more flexible electrode deactivation strategy. Here we present an explainable artificial intelligence (XAI) model fit on a large longitudinal dataset that can 1) predict at which point in time the manufacturer chose to deactivate an electrode as a function of routine clinical measures ("predictors") and 2) reveal which of these predictors were most important.

Methods : We analyzed a longitudinal dataset of 5496 perceptual thresholds and electrode impedances measured on 627 electrodes in 12 Argus II patients (Second Sight Medical Products). The data was collected from 2007-2018 during 285 sessions conducted at 7 different implant centers. To prepare the raw data for machine learning, we combined threshold and impedance values with clinical data crowd-sourced from the literature, and split the features into three different categories: 1) Routinely collected data (e.g., age at blindness onset, age at implant surgery), 2) System fitting (e.g., thresholds, impedances, charge density limits), and 3) Follow-up examinations (more recent threshold and impedance measurements). We then used gradient boosting (XGBoost), a powerful XAI model based on decision trees, to predict electrode deactivation as a function of these features.

Results : The model predicted electrode deactivation from routinely collected data with 60.8% accuracy. Performance increased to 75.3% with system fitting data, and to 84% when thresholds from follow-up examinations were available. The model further identified subject age and time since blindness onset as important predictors of electrode deactivation.

Conclusions : On the one hand, these findings highlight the importance of periodical threshold measurements to continuously monitor device performance. On the other hand, in the absence of such measurements, our work demonstrates that routinely collected clinical measures and a single session of system fitting might be sufficient to inform an XAI-based electrode deactivation strategy for retinal prostheses.

This is a 2021 ARVO Annual Meeting abstract.

×
×

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.

×