Nonparametric estimation of conditional ROC curves: Application to discrimination tasks in computerized detection of early breast cancer

A local linear method for estimating the conditional ROC curve under the presence of continuous and categorical covariates has been introduced in this paper. A data driven smoothing parameter has been proposed via the bootstrap method. The methods have been illustrated with real data coming from a discrimination problem emerging in the context of a computer-aided diagnosis system. The bootstrap approach has been also used in this setup to construct pointwise confidence intervals for the area under the ROC curve.