Variable Selection and Classification Using Computed Tomography (CT) Medical Image Data

• Activity Number: 534
• Type: Topic Contributed
• Date/Time: Thursday, August 2, 2007 : 10:30 AM to 12:20 PM
• Sponsor: IMS
• Author(s): Hyun Kim*+ and Gang Li and David W. Gertson and Robert Ochs and Matthew S. Brown and Jonathan Goldin
• Companies: University of California, Los Angeles and University of California, Los Angeles and University of California, Los Angeles and University of California, Los Angeles and University of California, Los Angeles and University of California, Los Angeles
• Keywords: variable selection ; classification ; Computed Tomography ; high dimension ; texture ; image

Abstract

PURPOSE: CT data is high dimensional and spatially correlated. The purpose of this study is to evaluate the sensitivity of two classification methods, namely, multinomial logistic regressions using backward selection and non-concave penalized likelihood feature selection using support vector machine classification for lung disease using texture features. METHOD: Each method was evaluated with and without clustering. CONCLUSIONS: For the multinomial model the classification accuracy was 91.4% and 91.1% with and without subject cluster respectively. The SVM classification rate was 91.7%. 40 features were commonly selected. In textural classification of lung diseases, accounting for possible intra-subject dependencies in the training set does not affect the resulting classification on a separate test set.