Early pigmentary retinosis diagnostic based on classification trees

Abstract

In this work we analyze different classification tree based techniques (CART, Bagging and Boosting), evaluating their performance with respect to their capability to reduce error rate and correct pattern classification. As a case of study we propose the classification of Pigmentary Restinosis patients through electroretinograms. Pigmentary Restinosis is the most frequent retina dystrophy (1/5000). The electroretinogram (ERG) constitutes a fundamental test in the study of this type of dystrophy since the wide clinical heterogeneity of visual diseases. Besides, retina electrophysiological study can provided information that may be used to predict the disease before the apparition of symptoms and allows us to corroborate the affectation degree on the dystrophic process of cones and canes. As experimental database we use a set of 148 electroretinograms , which is part of a retrospective study carried out by the Cuban National Reference Center of Pigmentary Retinosis. © 2011 Springer-Verlag.