Paired receiver operating characteristic curves and the effect of history on radiographic interpretation. CT of the head as a case study

Abstract

The use of a statistical technique for paired comparisons using receiver operating characteristic (ROC) curves is illustrated by studying the extent to which clinical history altered the interpretation of computed tomographic (CT) examinations of the head. Eighty-nine CT examinations of the head were presented in random order to four readers, first with minimum history (age and sex) and then several weeks later with complete neurological history as of the time the CT examination had been obtained. Using a paired ROC analysis, a small but significant (p less than .05) improvement was detected for the interpretations in the presence of complete history; for readings without history the average area was 94.4% and for readings with history it was 97.7%.