Evaluation of GDx parameters by using information theory Bilgi kurami kullanilarak GDx parametrelerinin deǧerlendirilmesi


Arslan U., Bozkurt B., Karaaǧaoǧlu A. E., Irkec M. T.

Turkish Journal of Medical Sciences, vol.41, no.1, pp.117-124, 2011 (SCI-Expanded) identifier

  • Publication Type: Article / Article
  • Volume: 41 Issue: 1
  • Publication Date: 2011
  • Doi Number: 10.3906/sag-0909-284
  • Journal Name: Turkish Journal of Medical Sciences
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, TR DİZİN (ULAKBİM)
  • Page Numbers: pp.117-124
  • Keywords: Diagnostic test, Glaucoma, Information theory, Roc curve
  • Lokman Hekim University Affiliated: No

Abstract

Aim: To evaluate the performance of GDx parameters in the diagnosis of glaucoma by using information theory and compare the results obtained using receiver operating characteristic (ROC) curve analysis, which is a traditional method. Materials and methods: Retinal nerve fiber layer thickness was measured in 270 eyes with glaucoma and 81 normal eyes with scanning laser polarimeter (NFA GDx version, 1.0.08), and 14 GDx parameters were calculated. Both ROC curve analysis and information theory were used to determine the best GDx parameters. The best cut-off points of these parameters were obtained using information theory for glaucoma prevalence (Pr) of 1%, 2%, and 5%. Results: The parameters having the maximum information content and discriminatory power are The Number, Ellipse modulation, and Maximum modulation, respectively. The best cut-off points associated with these parameters are 32.08, 1.65, and 1.20 for specified Pr values considered, respectively. The best cut-off value for inferior ratio is 1.95 when Pr is 1% or 2%, whereas the best cut-off point of the parameter is 2.11 when Pr is 5%. Conclusion: Although ROC curve analysis can be used for evaluating the performance of the diagnostic test, it cannot determine the best cut-off point for certain prevalence. Information theory approach seems to be more superior to the traditional ROC curve analysis for tackling this problem. © TÜBİTAK.