Local and Non-local Regularization Techniques in Emission (PET/SPECT) Tomographic Image Reconstruction Methods.

Emission tomographic image reconstruction is an ill-posed problem due to limited and noisy data and various image-degrading effects affecting the data and leads to noisy reconstructions. Explicit regularization, through iterative reconstruction methods, is considered better to compensate for reconstruction-based noise. Local smoothing and edge-preserving regularization methods can reduce reconstruction-based noise. However, these methods produce overly smoothed images or blocky artefacts in the final image because they can only exploit local image properties. Recently, non-local regularization techniques have been introduced, to overcome these problems, by incorporating geometrical global continuity and connectivity present in the objective image. These techniques can overcome drawbacks of local regularization methods; however, they also have certain limitations, such as choice of the regularization function, neighbourhood size or calibration of several empirical parameters involved. This work compares different local and non-local regularization techniques used in emission tomographic imaging in general and emission computed tomography in specific for improved quality of the resultant images.

NON-LOCAL PRIORS FOR HIGH-DIMENSIONAL ESTIMATION.

Jointly achieving parsimony and good predictive power in high dimensions is a main challenge in statistics. Non-local priors (NLPs) possess appealing properties for model choice, but their use for estimation has not been studied in detail. We show that for regular models NLP-based Bayesian model averaging (BMA) shrink spurious parameters either at fast polynomial or quasi-exponential rates as the sample size n increases, while non-spurious parameter estimates are not shrunk. We extend some results to linear models with dimension p growing with n. Coupled with our theoretical investigations, we outline the constructive representation of NLPs as mixtures of truncated distributions that enables simple posterior sampling and extending NLPs beyond previous proposals. Our results show notable high-dimensional estimation for linear models with p ≫ n at low computational cost. NLPs provided lower estimation error than benchmark and hyper-g priors, SCAD and LASSO in simulations, and in gene expression data achieved higher cross-validated R2 with less predictors. Remarkably, these results were obtained without pre-screening variables. Our findings contribute to the debate of whether different priors should be used for estimation and model selection, showing that selection priors may actually be desirable for high-dimensional estimation.