Accuracy of channelized hotelling observer performance measures estimated from repeated CT scans

We consider estimation of channelized Hotelling observer performance measures from a finite sample of reconstructed CT images, resulting from repeated scans. After observing that precise statistical theory for estimates of the SNR and AUC does not appear to be well-known in the medical imaging literature, we demonstrate how known results from the statistics literature may be applied to better understand the distributions of estimates for the SNR and AUC. In addition, we provide exact expressions for the bias and variance of the estimated SNR² and show how the moments for the estimated AUC may be numerically evaluated, without Monte Carlo simulation.