Optimum compensation method and filter cutoff frequency in myocardial SPECT: A human observer study

Attenuation, photon scatter, and distance-dependent collimator-detector response are major degrading factors in myocardial SPECT images. The current study investigated whether compensation for these factors improves perfusion defect detectability, and compared the results for human observers with a previous study using a mathematical observer. Methods: Four methods were investigated: attenuation compensation (AC); attenuation and detector response compensation; attenuation and scatter compensation; and attenuation, detector response, and scatter compensation (ADSC). For ADSC, 4 three-dimensional postreconstruction Butterworth filter cutoff frequencies were investigated for a pixel size of 0.62 cm: 0.12, 0.14, 0.16, and 0.22 pixel^-1. Five observers read images reconstructed using the 4 compensation methods. Receiver operating characteristics (ROC) analysis was used to determine the area under the ROC curve in each treatment studied. Results: Reconstruction methods that incorporated scatter and detector response compensation had higher indices of detectability than AC alone. Over the range studied, a filter cutoff frequency of 0.14 pixel^-1 was optimal. A comparison of human observer results with an earlier channelized Hotelling observer study performed with the same images showed excellent agreement in trend and ranking of defect detectability. Conclusion: Compensation for detector response and scatter improves defect detectability compared with AC alone, although detectability may depend on phantom population choice and noise level. An optimal filter cutoff was found that is lower than what is typically used in a clinical setting. The channelized Hotelling observer is a good predictor of human observer performance and may reduce the need for tedious, time-consuming studies with human observers.