A Family of Nonparametric Statistics for LROC Curves

The location response operating characteristic (LROC) curve provides an elegant tool to evaluate observer performance in both correctly localizing a lesion in an abnormal image and distinguishing an abnormal image from a normal image, while the receiver operating characteristic (ROC) curve is only able to evaluate an observer´s performance of the latter task. At the beginning of this paper, we review a general family of nonparametric summary statistics for ROC curves. After we discuss a parametric LROC method, we propose an alternative nonparametric method to estimate the area under the LROC curve. Our estimator consists of a general family of nonparametric summary statistics for LROC curves. We illustrated the use of our estimator through a diagnostic imaging example.