Estimation of trained-observer performance with known difference of class means

Abstract

This work concerns ROC estimation for task-based image quality assessments involving binary discrimination tasks. We investigate the statistical advantage that may be gained in ROC estimates by assuming that the difference of the class means for the observer ratings is known. Such knowledge can be obtained, for example, in image quality studies employing linear model observers and known-location lesion detection tasks with images reconstructed from either simulated data or real data collected using phantoms. To carry out this investigation, we introduce parametric point and confidence interval estimators under two scenarios: (1) the class means are both known, and (2) the difference of class means is known. An evaluation of our new estimators for the area under the ROC curve establishes that a large reduction in statistical variability can be achieved by using knowledge of the difference of class means. We demonstrate the usefulness of our approach with an image quality assessment example using real CT images of a thorax phantom.