Automatic classification of text documents presenting radiology examinations

Abstract

The paper presents the classification of text documents presenting radiology examinations, taking into consideration two groups: cases with aneurysms and those without it. A database containing descriptions of 1284 cases was classified using the maximum entropy algorithm and frequent phrase extraction. It was revealed that the best method was the classifier using the maximum entropy algorithm based on nouns. The best result obtained was 90% of sensitivity and 70% of specificity. The worse diagnostic capacity demonstrates frequent phrase extraction algorithm. The other classifiers turned out to be less effective, than the random ones.