Programm & Abstracts                 "Innovationen in der Augenheilkunde"

Aktuelle Tagungsinformationen
   News and Updates

Anmeldung zur Tagung
   Registration
Hotelbuchung
   Hotel Registration
Grußwort
   Welcome address
Beteiligte Gesellschaften
   Societies involved
Eröffnung des Kongresses
   Opening Ceremony
Preise
   Awards
Wissenschaftliches Programm
   Scientific program
Posterpräsentationen
   Poster Presentation
Kurse
   Courses
Begleitende Veranstaltungen
   Collateral Events
Rahmenprogramm
   Social program
Jubiläumsparty
   Jubilee Party
DOG Information
   DOG Information
Allgemeine Informationen
   General Information
Autorenindex
   Index of Authors
Ausstellerliste
   Exhibitors
Sponsoren
   Sponsors
Teilnahmegebühren
   Registration fees
Impressum



DOG Homepage

Diagnosis of Glaucoma by Indirect Classifiers

1Peters A., 1Lausen B., 2Michelson G., 1Tessmar J., 1Gefeller O.,
1Friedrich-Alexander-Universität, Institut für Medizininformatik, Biometrie und Epidemiologie (IMBE) (Erlangen)
2Friedrich-Alexander-Universität, Augenklinik (Erlangen)

Purpose: To classify subjects as normal or glaucomatous based on data from the Heidelberg Retina Tomograph, papillometry, perimetry and medical history.
Methods: Data from a cross-sectional study including 85 glaucomatous and 85 normal eyes (matched by age and sex) are classified to normal or glaucoma. A framework called indirect classification (cf. Hand et al., 2001) is used to combine medical knowledge about the disease and statistical classification methods in order to reduce the misclassification error, i.e. to increase sensitivity and specifity. Classification trees are the utilized statistical tool in the framework of indirect classification. Bagging the procedure leads to a reduction of the misclassification rate, cf. Breiman (1996). Indirect classification results are compared with results of linear discriminant analysis, classification trees and bagged classification trees.
Results: Linear discriminant analysis classifies 29.0% (Senitivity: 71.0%, Specifity: 75.5%) of the eyes incorrect. The misclassification error is 27.3% (Senitivity: 72.5%, Specifity: 72.9%) using classification trees and lowered onto 19.3% (Senitivity: 79.7%, Specifity: 81.7%) by bagging the classification trees. In contrast to that, indirect classification achieves in general smaller misclassification results. Incorporating classification trees in the framework of indirect classification classifies 22.8% (Senitivity: 78.5%, Specifity: 75.9%) of the eyes incorrect. Bagging the procedure lowers the misclassification rate onto 17.9% (Senitivity: 87.3%, Specifity: 76.9%).
Conclusions: Classification based on the given data is improved by combining medical knowledge and statistical classification methods.
References: Breiman L (1996): Bagging predictors, Machine Learning 24,123-140. Hand DJ, Li HG, Adams NM (2001): Supervised classification with structured class definitions, Computational Statistics & Data Analysis, 36, 209-225.

Zurück/Back