Comparison of radially-symmetric versus oriented channel. Models using channelized hotelling observers for myocardial defect detection in parallel-hole SPECT

The objective of this study was to compare defect detection performance found using radially-symmetric (RS) and radially-symmetric broken into oriented channels (OC) models in channelized Hotelling observers with that from a previously reported human observer ROC study. The main goal of the human observer study was to compare detection performance of left-ventricular myocardial perfusion defects using filtered-backprojection (FB) image reconstruction without attenuation compensation to defect detection performance using maximum-likelihood expectation-maximization (ML-EM) image reconstruction with nonuniform attenuation compensation (AC). This was performed for cases of no significant attenuation artifacts (male anatomy, flat diaphragm MF) and significant attenuation reconstruction artifacts (male with raised diaphragm, MR; female with large breasts, F). The areas under the ROC curves (A_x) for the Hotelling observers with both channel models (RS, OC) were calculated and compared to those found in the human observer study. Similar trends with the RC channel model were seen in average observer performance for both human and channelized Hotelling observers in both FB without AC and ML-EM with AC over the three different anatomy types except for the MF anatomy reconstructed with FB. The correlation coefficient for the 6 average observer performances was 0.84. Improved agreement between the human and channelized Hotelling observers was found when using the OC channel model. The correlation coefficient found was 0.94, again with the MF anatomy reconstructed with FB as the most significant difference between the human and channelized Hotelling observer results. Further investigation of the channel models is needed to determine number and alignment of orientations and number and cutoff of frequency bands to improve agreement between human and Hotelling observer results