Improvement of Glaucoma Diagnosis by Double-Bagging

Hothorn T., Lausen B.,
Friedrich-Alexander-Universität, Institut für Medizininformatik, Biometrie und Epidemiologie (IMBE) (Erlangen)

Purpose: It is the aim to classify subjects as normal or glaucomatous based on laser scanning data from Heidelberg Retina Tomograph (HRT) and anamnestic data using a combination of linear and tree classifiers.
Method: Bootstrap aggregation of classification trees lead to improved error rates for glaucoma classification based on standard HRT variables (Mardin et al., 2002). Monte-Carlo studies on classifiers for HRT variables indicate that bagged classification trees perform well when subgroups of patients with different characteristics are investigated. Based on this results we suggest a combination of both LDA and classification trees ("Double-Bagging") for further improvements in glaucoma classification.
Results: Double-Bagging performs comparable to the best of both methods. For the cross-sectional study (Mardin et al, 2002), the estimated misclassification error of Double-Bagging is 14.3% (Sensitivity: 81.6%, Specifity: 89.8%).
Conclusions: The use of the combined classifiers decreases error rates for glaucoma classification and the estimated error rate is unbiased for method selection.
References: Mardin CY, Hothorn T, Peters A, Jünemann AG, Michelson G, Lausen B (2002): New Glaucoma Classification Method Based on Standard HRT Parameters by Bagging Classification Trees, submitted

Copyright © Deutsche Ophthalmologische Gesellschaft, 2002. All rights reserved.

Last updated: 

Webdesign: SPALLEK.COM