Comparison of LROC and Traditional ROC Studies for Lesion-Detection Task

The goal of this study is to evaluate lesion detectability in In-111 ProstaScintreg SPECT with different image reconstruction methods using receiver operating characteristic (ROC) techniques of traditional signal position known exactly (SPKE) settings and two other different settings. We generated 3D NCAT phantoms with different body sizes, organ uptake ratios, lesion sizes and lesion contrasts to simulate variations found in a patient population. Monte Carlo methods were then used to generate projection data. The projections were reconstructed using the OS-EM algorithm with different compensations including attenuation compensation (AC), AC and scatter compensation (SC) and AC, SC and collimator-detector response compensation (CDR). We used the channelized Hotelling observer (CHO) for all the 3 settings of ROC studies. The area under ROC curves (AUC) was used as the figure of merit to optimize the updates number and postfilter cut-off frequency of OS-EM with different compensations and then compare the optimized methods. The results show that all 3 different settings of ROC study give consistent results for comparing optimized OS-EM with different compensations. ROC setting 2, which involves localization uncertainty and nonlinear combination of body size variations have better statistical power than the SPKE setting (setting 1). The AUC of CHO in ROC setting 2 and 3 are significantly lower than that of setting 1 which may predict better the human observers´ performances in clinical trials.