Investigative Ophthalmology & Visual Science Cover Image for Volume 61, Issue 7
June 2020
Volume 61, Issue 7
Free
ARVO Annual Meeting Abstract  |   June 2020
An artificial intelligence app for strabismus
Author Affiliations & Notes
  • Laura Alves de Figueiredo
    Strabismus, Hospital das Clínicas da Faculdade de Medicina da USP (HC FMUSP), São Paulo, São Paulo, Brazil
  • Iara Debert
    Strabismus, Hospital das Clínicas da Faculdade de Medicina da USP (HC FMUSP), São Paulo, São Paulo, Brazil
  • João Victor Pacheco Dias
    Strabismus, Hospital das Clínicas da Faculdade de Medicina da USP (HC FMUSP), São Paulo, São Paulo, Brazil
  • Mariza Polati
    Strabismus, Hospital das Clínicas da Faculdade de Medicina da USP (HC FMUSP), São Paulo, São Paulo, Brazil
  • Footnotes
    Commercial Relationships   Laura Figueiredo, None; Iara Debert, None; João Victor Dias, None; Mariza Polati, None
  • Footnotes
    Support  None
Investigative Ophthalmology & Visual Science June 2020, Vol.61, 2129. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Laura Alves de Figueiredo, Iara Debert, João Victor Pacheco Dias, Mariza Polati; An artificial intelligence app for strabismus. Invest. Ophthalmol. Vis. Sci. 2020;61(7):2129.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : Clinical evaluation of eye versions is important in the diagnosis of patients with uncommon strabismus and restricted eye movements. Despite the importance of versions, there is no standardization in clinical practice as these are subjective evaluation subjected to bias. Based on the assumption that objectivity confers accuracy,this research aimed to create an artificial intelligence app capable of locating and classifying the versions in the nine positions of gaze.

Methods : We analyzed photos of 105 strabismus patients from outpatient clinic of a tertiary hospital at nine gazes from 2015 to 2019. For each photo there was a gaze identification and the corresponding version rating by the same examiner during the patient examination. Data were quantified by descriptive statistical analysis from 1 to 9 in reference to the gaze and from –4 to +4 in reference to the version classification.

Results : Practicality guided the options, the first step was selecting the operating system for mobiles to be the platform, and the development environment was built in the Phython programming language as base-language and the Keras as photo framework for modeling. During the standardization of images, landscape and portrait aspects were corrected. A face detection library was used to obtain their face center and thereby generate an image crop for each eye. These two images from each patient were used to generate, through a data augmentation process, 34 new images each. Some of the modifications were Brightness, Rotation, RGB Shift,Random Gamma and Optical Distortion. These 6000+ images were used as fine-tuning in a pre-trained Neural Network model with the Transfer Learning technique. For validation, besides using k-folds was also applied test-time augmentation. The final model was exported for smartphone use via Tensorflow Mobile converter. As a consequence, the mobile app was capable to recognize the position of eye in the nine gazes and the respective version classification of each eye.

Conclusions : The results showed that the applicability of the mobile app to the ophthalmic routine might complement the evaluation of ocular motricity based on the objective classification of the ocular versions. Relative to the traditional method of clinical practice, professionals will be able to envisage the possibility of an easy-to-apply support app,as well as reducing time and increasing diagnostic accuracy. However,further exploratory research and validation are required.

This is a 2020 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.

×