Predicting the variance of ML reconstructions with body contour constraint for multi-pinhole SPECT

A well known limitation of the multi-pinhole system is that the reconstructed images may suffer from artifacts due to truncated and/or overlapping projections. A previous study showed that these artifacts can be strongly suppressed by using prior knowledge about the object boundary, which is typically obtained from a CT- or MRI-scan acquired in the same pose. Forcing the reconstructed activity outside the object contour to zero effectively eliminated the artifacts. In this study, we investigate the validity of an analytical image quality prediction method, which previously has been used to optimize the multi-pinhole system, with and without the application of body contour constraint in the post-smoothed MLEM reconstruction. The results indicate that the use of the body contour constraint has hardly noticeable effects on the variance, which is well predicted by the analytical method. In contrast, the contour constraint eliminated the bias and artifacts as expected. The analytical method assumes an unbiased estimator, and therefore, it is a good predictor for post-smoothed MLEM with body contour constraint. Because the information about the body contour is often available in small animal imaging, we conclude that it seems valid to use the analytical method for optimizing the design of multi-pinhole collimators.