Use of cross-validation as a stopping rule in emission tomography image reconstruction

Simulated and real Positron Emission Tomography (PET) images are reconstructed using the iterative EM algorithm. The data is split up into independent parts. The EM algorithm is applied to each part and stopped according to a cross-validation procedure. For a variety of simulated and real data sets, stopping points were reached. For simulated data, the average of the reconstructions from a four-way split of the data was visually superior to the reconstruction obtained by applying the EM algorithm to the full data and stopping at a subjectively chosen iteration. For real data, such an improvement was not observed. To remove point artifacts, the reconstructions were filtered.