The Effects of Object Variability on the Channelized Hotelling Observer Performance in the Evaluation of R4SSH and PH Myocardial Perfusion SPECT

The purpose of this work is to investigate the effects of different kinds of object variability on the performance of the channelized Hotelling observer (CHO) for detecting defects in myocardial perfusion SPECT using rotating 4-segment slant-hole (R4SSH) and parallel-hole (PH) collimators. In this work, the categories of object variability used were: defect location, uptake ratio and anatomical variations. The effect of each kind of variation alone, the combination of two variations, and mixture of three variations were studied. R4SSH and PH projection data from the 3D NCAT phantom with radioactivity distribution modeling that of Tc-99m Sestamibi were generated using the SIMIND Monte Carlo code. Iterative OS-EM with 4 subsets was used to reconstruct the projections. The number of iteration and the cutoff frequency were optimized for each combination of object variability and imaging method in terms of area under the ROC curve (AUC). The results indicated that R4SSH SPECT allows significantly better myocardial defect detectability than that of PH SPECT when uptake ratio variation alone was included. The differences between defect detectability of R4SSH and PH SPECT became insignificant when defect location and anatomical variations were included in the population. We concluded that, when conducting observer studies to compare different imaging methods, including defect and anatomical variation in the phantom will reduce the statistical power. In clinic, diagnostic accuracy may be improved if radiologists stratify their reading as much as possible.