Local-mean preserving post-processing for non-negativity enforcement in 90Y-PET imaging

PET reconstruction algorithms allowing negative values in the image space have gained interest for twenty years [1] , [2] . They allow to solve problems introduced by conventional algorithms satisfying the non-negativity constraint such as MLEM. This may present a problem when low activity regions are involved such as those typically found in 90Y-PET. In such cases, algorithms allowing negative values in the image space can reduce the positive bias but often at the cost of a variance increase. A post-processing reducing the variance while keeping the reduction of the bias is therefore desirable. The present work proposes a new approach to reach this goal.