Maximum-likelihood estimation of normalisation factors for PET

In this work an iterative ML technique is developed to normalise acquired PET data. We proposed a model for component-based correction featuring geometric, crystal efficiency and block timing factors. The algorithm is tested against the conventional fan-sum method and with a non-ML iterative technique on both simulated and acquired data. The results show that the iterative methods are superior to the conventional fan-sum technique. Furthermore the new method provides an improved normalisation over the previously published iterative technique when low statistics acquisitions are used.