Evaluating attenuation correction in Ga-67 SPECT image reconstruction through numerical observer ROC and human observer LROC

The channelized Hotelling observer (CHO) has correlated with human observers for many “signal-known-exactly” detection tasks. The authors have found that average CHO performances in SKE ROC studies, taken over multiple tumor locations, can be used to eliminate some suboptimal reconstruction strategies from consideration in human localization ROC (LROC) studies. Here, the authors have continued this line of research using a human LROC study that evaluated attenuation correction (AC) for OSEM and FBP reconstruction. The detection task was “hot” tumor detection in simulated Ga-67 scans of the chest, and 3 tumor activities were used. The projection data modeled attenuation and detector blur, but perfect scatter rejection was assumed. Images were obtained from 5 reconstruction strategies: OSEM with AC, OSEM without AC, FBP without AC, FBP with multiplicative Chang AC, and FBP with 1 iteration of iterative Chang AC. The OSEM strategies used 1 iteration with 8 subsets of 16 angular projections each. Results in the human observer LROC study were based on 200 images (100 tumor locations) per strategy, while the CHO was applied with 100 images at each of 50 tumor locations. In both studies, statistically significant differences (p=0.05) were found between OSEM with AC and the other 4 strategies, but no other significant differences between strategies were found. Future investigations will examine analytical methods of obtaining image statistics for the CHO in order to make application of the CHO more efficient