System and reconstruction optimization in SPECT using model observers for different tasks

In SPECT the collimator and reconstruction algorithm both control image quality. For a given task, one might naturally seek that combination of collimator and algorithm parameters that maximizes task performance. Here, we use a brute force method to compute task performance over a joint space of collimator and algorithm parameters. Our interest is the task. We perform our brute force computation for a task involving detection and localization of a signal, and for a task involving detection only. We use human emulating model observers to carry out the tasks. The optimal parameters found by the detection-localization task differ from those of the detection-only task, and the sensitivity to parameter choice is much higher for the detection + localization task.