Investigative Ophthalmology & Visual Science Cover Image for Volume 59, Issue 9
July 2018
Volume 59, Issue 9
Open Access
ARVO Annual Meeting Abstract  |   July 2018
Convolutional Neural Networks for identification and classification of optic nerve damage features
Author Affiliations & Notes
  • Sandra Belalcazar
    Fundacion Oftalmologica Nacional, Bogota, Colombia
    Universidad del Rosario, Bogota, Colombia
  • Hernan Andres Rios
    Fundacion Oftalmologica Nacional, Bogota, Colombia
    Universidad del Rosario, Bogota, Colombia
  • Vanessa Carpio
    Fundacion Oftalmologica Nacional, Bogota, Colombia
    Universidad del Rosario, Bogota, Colombia
  • Oscar Julian Perdomo
    Universidad Nacional, Bogota, Colombia
  • Claudia Rosa Carvajal
    Fundacion Oftalmologica Nacional, Bogota, Colombia
  • Fabio A Gonzalez
    Universidad Nacional, Bogota, Colombia
  • Henning Müller
    University of Applied Sciences Western Switzerland HES-SO, Sierre, Switzerland
  • Footnotes
    Commercial Relationships   Sandra Belalcazar, None; Hernan Rios, None; Vanessa Carpio, None; Oscar Perdomo, None; Claudia Carvajal, None; Fabio Gonzalez, None; Henning Müller, None
  • Footnotes
    Support  None
Investigative Ophthalmology & Visual Science July 2018, Vol.59, 1719. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Sandra Belalcazar, Hernan Andres Rios, Vanessa Carpio, Oscar Julian Perdomo, Claudia Rosa Carvajal, Fabio A Gonzalez, Henning Müller; Convolutional Neural Networks for identification and classification of optic nerve damage features. Invest. Ophthalmol. Vis. Sci. 2018;59(9):1719.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : Optic nerve damage is one of the main features for identification of Glaucoma. The convolutional neural networks are a promising tool for early detection of several diseases using medical images, with great importance in developing countries, where access to medical services is difficult. The purpose of the study was to determine the accuracy of convolutional neural networks to identify early optic nerve damage features on optic nerve photographs as a method for automatic detection of Glaucoma suspects

Methods : A cross-sectional diagnostic test study was made. A designed model based on convolutional neural networks was used to analyze optic nerve photographs and automatically segment the optic disc, measuring the total optic disc diameter, the thinnest ring radius and the ratio between these two parameters to estimate the optic nerve damage

Results : The experimental results shown that the proposed model is able to calculate the total diameter, the thinnest ring radius and the ratio with a mean absolute percentage error of 9.9%, 23.9% and 24.8% respectively

Conclusions : Convolutional neural networks show good performance in detection of early optic nerve damage features. This study is the first step to build a telemedicine tool to support physicians to detect glaucoma suspects using optic nerve photographs

This is an abstract that was submitted for the 2018 ARVO Annual Meeting, held in Honolulu, Hawaii, April 29 - May 3, 2018.

 

Boxplot of model predictions (Blue = Convolutional Neural Network) versus measure taken by expert (Green = Ophthalmologist) about optic nerve total diameter

Boxplot of model predictions (Blue = Convolutional Neural Network) versus measure taken by expert (Green = Ophthalmologist) about optic nerve total diameter

 

Boxplot of model predictions (Blue = Convolutional Neural Network) versus measure taken by expert (Green = Ophthalmologist) about optic nerve thinnest ring radius

Boxplot of model predictions (Blue = Convolutional Neural Network) versus measure taken by expert (Green = Ophthalmologist) about optic nerve thinnest ring radius

×
×

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.

×