A comparison of human and model observers in multislice LROC studies

The performances of human and model observers in LROC studies involving a multislice image display have been compared. The task was detection of Ga-67 tumors within the MCAT phantom. Projection data for the 93- and 185-keV gallium emissions were generated with an analytic projector modeling nonuniform attenuation and collimator response. The observer studies evaluated four reconstruction strategies: the filtered-backprojection (FBP) and ordered-subset expectation-maximization (OSEM) algorithms with 2D and 3D postfiltering. Attenuation correction was performed for all strategies. Two image-display formats were used in the studies. One was a single-slice format, with which the observer scored an image based on that image alone. The second was a multislice format presenting a central slice and 4 neighboring slices that were to be used to score the central slice. The human-observer data was drawn from studies performed by Wells et al. [TNS, 2000]. Multiclass versions of the nonprewhitening (NPW), channelized nonprewhitening (CNPW), and channelized Hotelling (CH) observers, capable of performing the tumor search task, were applied. The channelized observers were tested with 3 square-profile channel models that were defined by the low-frequency cut-off ρ₀. With either display format, the best agreement with the human-observer performance trends was had with the CH observer with ρ₀=0.015 pixel^-1. However, quantitative agreement with the human performances requires incorporation of an internal-noise model.