Many of the investigations that are now carried out during imaging systems' evaluations and comparisons include interpretation (reading) studies of normal and abnormal cases that are evaluated by receiver operating characteristic analyses (ROC) or derivatives thereof. Unfortunately, limitations in present methodology for systems evaluations have results in extremely complicated and costly study designs and data methodology to apply a different formulation of a well- known statistic to this problem and verify the proposed approach using both simulation and a large set of experimental data available to us. The Wilcoxon statistic, a non-parametric method often used to obtain the area under the ROC curve when scoring data is used, has a formulation that compares all possible pairs of observations of a negative and a positive case. Although this conceptualization of the problem can be used to further develop and extend ROC methodology. We plan to exploit this conceptualization to develop a sensitive statistical test to compare the areas under two or more ROC curses in a manner that will enable us to assess the effect of selected variables on the area under the ROC curve as to summarize diagnostic accuracy in the multi-disease setting and to quality the difference in diagnostic accuracy of two modalities (or more) as a function of different parameters and possibly after the incorporation of utility function.
Showing the most recent 10 out of 11 publications
