April 2014
Volume 55, Issue 13
Free
ARVO Annual Meeting Abstract  |   April 2014
Method to detect visual field loss using multifocal visual evoked potential
Author Affiliations & Notes
  • Givago S Souza
    Insituto de Ciencias Biologicas, Universidade Federal do Para, Belem, Brazil
    Nucleo do Doenças Tropicais, Universidade Federal do Para, Belém, Brazil
  • Iraquitan Cordeiro Filho
    Instituto Tecnologico VALE, Belem, Brazil
    Instituto de Ciencias Exatas e Naturais, Universidade Federal do Para, Belem, Brazil
  • Lorena Botelho Vergara
    Instituto de Ciências da Saúde, Universidade Federal do Para, Belem, Brazil
  • Alexandre Rosa
    Instituto de Ciências da Saúde, Universidade Federal do Para, Belem, Brazil
  • Hideraldo Cabeça
    Hospital do Estado Ophir Loyola, Belem, Brazil
  • Schubert Carvalho
    Instituto Tecnologico VALE, Belem, Brazil
  • Footnotes
    Commercial Relationships Givago Souza, None; Iraquitan Cordeiro Filho, None; Lorena Botelho Vergara, None; Alexandre Rosa, None; Hideraldo Cabeça, None; Schubert Carvalho, None
  • Footnotes
    Support None
Investigative Ophthalmology & Visual Science April 2014, Vol.55, 5127. doi:https://doi.org/
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Givago S Souza, Iraquitan Cordeiro Filho, Lorena Botelho Vergara, Alexandre Rosa, Hideraldo Cabeça, Schubert Carvalho; Method to detect visual field loss using multifocal visual evoked potential. Invest. Ophthalmol. Vis. Sci. 2014;55(13):5127. doi: https://doi.org/.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose: To develop a supervised learning model with decision trees to act as a decision support system in the task to identify visual field losses by using multifocal visual evoked cortical potential (mfVECP).

Methods: We studied mfVECP data from 22 eyes of healthy subjects, 23 eyes of subjects with neuromyelitis optica (NMO), and 16 eyes from subjects with multiple sclerosis (MS). Dartboards with 60 checkboard sectors were used as stimuli. The SNR from the waveforms of the 60 sectors was used as input into a decision tree. A combination of two-sample t-test feature selection and forward sequential feature selection was used to obtain the best representative features or sectors that sorted out the three subject classes. With these selected features, decision tree models were fitted and cross-validated using leave-one-out technique. For evaluation criteria, we used the classification of accuracy and Cohen’s Kappa coefficient. The following comparisons were analyzed: healthy subjects versus NMO subjects; healthy subjects versus MS subjects; NMO subjects versus MS subjects; healthy subjects versus subjects suffering for both illnesses.

Results: We found a number of visual field sectors that better differentiate mfVEP data for each comparison: 12 sectors (healthy subjects vs NMO subjects); 24 sectors (healthy subjects vs MS subjects); 21 sectors (healthy subjects vs both illnesses); none sector (NMO subjects vs MS subjects). With the selected features, decision trees were fitted following the same combinations of the feature selection process. For each combination, two decision trees were fitted, one without the feature selection process and another with the selected features. In the healthy vs NMO dataset the results after cross-validation shows that the accuracy of the decision tree and the Kappa value were 80% and 0.6002, respectively; for this same dataset after the feature selection the results were 84% and 0.6884, respectively. For healthy vs MS dataset, the accuracy and Kappa value were 71.05% and 0.4011, respectively (no feature selection), and after feature selection the results were 76% and 0.5100, respectively. For healthy vs both illnesses as one class, the accuracy and Kappa value were 68.85% and 0.3039, respectively, after feature selection the results were 68.85% and 0.3178, respectively.

Conclusions: In general, the use of feature selection improved classification accuracy.

Keywords: 507 electrophysiology: clinical • 612 neuro-ophthalmology: diagnosis • 755 visual cortex  
×
×

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.

×