ROI level analysis may involve clustering of lesions (e.g lung nodules) in a particular ROI, leading to erroneous sensitivity values if usual statistical analysis is done. Analysis must account for the clustering. In these cases, how do you prepare ROC curves?