Comparison of Multi-slice CHO-HO Models and Human Observers in Defect Detection with SPECT Myocardial Images
The objective of this study was to compare defect detection performance using a multi-slice CHO-HO (Channelized Hotelling Observer - Hotelling Observer) model with performance found in two previously reported human observer receiver operating characteristic (ROC) studies. The tasks were left-ventricular myocardial perfusion lesion detection, with simulated projection data reconstructed using ordered subsets expectation maximization (OSEM) without attenuation and scatter correction (ASC) (ROC study one), and with experimental phantom data reconstructed using OSEM with ASC (ROC study two). Both simulated and experimental data modeled female and male anatomy. The projections were setup to simulate equal time dual-head anterior 180° and triple-head 360° scans. The multi-slice lesion detectability indices for the CHO-HO were calculated and compared to the A-'s for the human observers. Multiple channel setups for the CHO were investigated. Positive correlations were found between these results for some channel setups. Further investigation of channel models is needed to improve performance correlation with human observers.