International Society for Industrial Process Tomography

Simultaneous Compensation for Attenuation, Scatter and Detector Response for 2D-Emission Tomography on Nuclear Waste with Reduced Data

Raphaël Thierry1, Jean-Luc Pettier1, Laurent Desbat2 1Commissariat à L' Energie Atomique de Cadarache, DCC/DESD/SCCD/LEC 13108 -St-Paul Lez Durance, France,

Phone : (+33) 04-42-25-61-97, (+33) 04-42-25-49-19, Fax : (+33) 04-42-25-73-77,

E-mail : thierry@desdcad.cea.fr, pettier@desdcad.cea.fr

2TIMC-IMAG, UMR CNRS 5525

Faculté de medecine, 38706 La Tronche Cedex Phone : (+33) 04-76-54-96-00, Fax : (+33) 04-76-54-95-55,

E-mail Laurent.Desbat@imag.fr

Abstract - We propose a simultaneous correction including three-dimensionnal detector/collimator response, attenuation and diffusion for characterization and appraisal of Nuclear Waste. An Emission Computed Tomography (ECT) imagery was developed in the framework of parallel-hole collimation associated with a HyperPure Germanium (HpGe) detection system. The procedure involves the evaluation of the analytical line spread function (LSF) model with the acquisition of the system LSF by sampling the detection kernel with a multi-energetic europium

(152Eu) linear source. We use a Maximum a Posteriori (MAP) estimator for the image restitution where the projection processing likelihood is either modeled using

a Poissonian or a Gaussian distribution.The introduction of the a priori distribution is chosen as to preserve the image discontinuities by the use of regularization. We compare a MAP-EM One Step Late (OSL) to a deterministic relaxation algorithm for edge-preserving regularization under specific 200-l mock-up and real packages. The analysis of the results showed an improved performance compared to other

models, both in quantification and in localization of radio-elements. On the whole

experiments it has been showed that volumic model is more precise in term of quantification than global measurements methods even within the latter's assumption framework. Considering the sources repartition, the volumic model globally reduces the blurring and the distortions in heterogeneous activity distributions phantoms. Further experimentions and simulations should give us more information on the influence of parameters, such as the bayesian hyperparameter or the measurement sampling configuration, over the accuracy of the reconstruction.

Keywords : Bayesian Reconstruction, Nonlinear Regularization, LSF Modelling, Nuclear Waste