ABSTRACT
This chapter describes the Obuchowski-Rockette (OR) method introduced in 1995, subsequently shown to be substantially equivalent to the DBMH method. Unlike the DBMH method, in the OR method the figure of merit is modeled directly. The two models have the same number of parameters, except that the covariance of the error term contains four terms, which account for different types of variances present in MRMC data. Since it holds the key to understanding the method, the error covariance is explained in detail by using the case-set {c} index. Resampling methods are used to estimate the error covariance terms. Described are random-reader random-case, fixed-reader random-case, and random-reader fixed-case analyses. Applications of the method using the capabilities of the RJafroc package are given. The corresponding code is brief as most of the complexity is cloaked in the package. An online appendix contains a longer but more transparent implementation that follows the formulae. Another online appendix describes single-modality multiple-reader analysis, which is used in Chapter 22 to compare CAD to a group of radiologists interpreting the same cases.
