Edge-preserving regularization in emission tomography with transmission CT boundary information: characterization of waste drums

Emission computed tomography (ECT) coupled with transmission computed tomography (TCT) is employed to evaluate the quantity and distribution of gamma radionuclides in heterogeneous radioactive waste packages. High-purity germanium (HPGe) detectors with good energy resolution are used but request long measuring time. Therefore, characterizing the whole package is carried out by reducing the measuring time and/or by under-sampling the data. In this paper, we investigate a maximum a posteriori (MAP) estimator using mixed constraints such as a binary activity support and an anisotropic gradient-based smoothing. The object support is deduced from a pre-segmentation of the attenuation map. The smoothing operator is locally calculated on a gradient constraint and a prior information on density edges. Besides, the MAP is expressed with a Poissonian likelihood and is maximized with a half-quadratic regularization which assures the positiveness of the solution. We compare it with a standard maximum likelihood (ML-EM) and a version (MOISE) that do not use the information on density edges. We did simulate the data of a 220-litre cylindrical drum, containing multi-energy gamma emitters. The results show that the MAP leads to significantly edge enhancement with a lower normalized mean square error (NMSE). The gains obtained in terms of resolution and quantification illustrate that boundary information could improve emission reconstruction for noisy and limited data as well as for low and high energy emitters.