Strategies to jointly optimize spect collimator and reconstruction parameters for a detection task

Abstract

In systems like SPECT, raw data is obtained by the imaging system and then reconstructed and viewed by a human observer. We compare two approaches to optimizing SPECT for a detection task with a known signal in a statistically varying background. In a sequential approach, we optimize the collimator using an ideal observer applied to the sinogram. We then optimize the regularization of the reconstruction using a human-emulating channelized Hotelling observer (CHO). In a second approach, we use the CHO to jointly optimize the collimator and regularization. The performance of the joint approach exceeds that of the sequential approach. The collimator properties from the joint approach are closer to that of a commercial collimator than those of the sequential approach. Thus using the “best” collimator derived by an ideal observer leads to suboptimal net detection performance.