A New Approach for Testing the Significance of Differences Between ROC Curves Measured from Correlated Data

摘要: Receiver Operating Characteristic analysis is now generally recognized as the most appropriate methodology for evaluating diagnostic performance of medical imaging procedures (1–7). ROC has been used in field psychophysics three decades, and its theory experimental have developed considerable detail (8–13). Perhaps surprisingly, statistical properties measures had received relatively little attention until several years ago, when limited size practical data sets applications indicated need careful study this issue. Recent progress includes work Metz Kronman (14,15), who a bivariate test significance differences between curves measured from independent sets; Hanley McNeil, studied area under an curve techniques to predict number cases required tc demonstrate “Area Indexes” either (16) or correlated (17) Swets Pickett (7), identified components variation outlined general protocol testing Area Index.