Task-based human observer study for evaluation and optimization of 3D & 4D image reconstruction methods for gated myocardial perfusion SPECT

Objectives The objective is to determine the optimal 4D image reconstruction method and associate parameters for gated myocardial perfusion (GMP) SPECT using a task-based human observer study.

Methods We previously evaluated the quality of GMP SPECT images from different 4D reconstruction methods in terms of trade-offs between normalized mean squared error (NMSE) and normalized standard deviation (NSD). In this study, we generated noise-free and noisy projection data with 16 cardiac gates from the 4D NCAT phantom with a realistic beating heart model with and without a regional myocardial motion defect (RMMD). An analytical projector that included the effects of attenuation, collimator-detector response and scatter (ADS) was used in the simulation. The projection data were reconstructed using 3D FBP and 3D OS-EM methods with ADS corrections followed by 4D filtering, and the 4D MAP-RBI-EM method with ADS corrections. The task of 5 observers including NM physicians was to rate their confidence in detecting the RMMD in the 4D GMP SPECT images. We applied ROC analysis and used the area under the ROC curve (AUC) to determine observer performance.

Results The optimal NMSE-NSD trade-off from optimized reconstruction parameters corresponds to maximum AUC value. The 4D MAP-RBI-EM which has the best NMSE-NSD trade-off among the methods also shows the highest AUC value.

Conclusions The NMSE-NSD trade-off was shown to agree with observer performance in detecting the RMMD in the 4D GMP SPECT images. The 4D MAP-RBI-EM with optimized parameters and ADS corrections was shown to provide the highest performance in the 4D motion defect detection task.