Cross-validation stopping rule for ML-EM reconstruction of dynamic PET series: Effect on image quality and quantitative accuracy

A major shortcoming of the maximum likelihood expectation maximization (ML- EM) method for reconstruction of dynamic positron emission tomography (PET) images is to decide when to stop the iterative process for image frames wi th largely different statistics and activity distributions. A widespread pr actice to overcome this problem involves overiteration of an image estimate followed by smoothing. In this paper, we investigate the qualitative and q uantitative accuracy of the cross-validation procedure (CV) as a stopping r ule, in comparison to overiteration and post-filtering, for the reconstruct ion of phantom and small animal dynamic F-18-fluorodeoxyglucose PET data ac quired in two-dimensional mode. The CV stopping rule ensured visually accep table image estimates with balanced resolution and noise characteristics. H owever, quantitative accuracy required some minimum number of counts per im age. The effect of the number of ML-EM iterations on time-activity curves a nd metabolic rates of glucose extracted from image series is discussed. A d ependence of the CV defined number of iterations on projection counts was f ound that simplifies reconstruction and reduces computation time.