-We present a method for obtaining attenuation maps for use in emission computed tomography (ECT) using ultra low dose CT data (at 140kVp, down to 10mA). This is achieved using a recursive k-means clustering method, the output of which initializes successive parameter-less region growing procedures. The method automatically produces templates corresponding to bone, lung, soft and dense tissue (muscle and fat). The segmentation of each tissue class from k-means clustering is used to compensate for the higher statistical noise variation seen at lower dose. The use of the region grower provides local contextual information that minimizes the impact of global noise. The templates were assigned appropriate linear attenuation coefficients and then convolved with the PET/SPECT system's PSF. This approach was applied to a dataset from an experimental anthropomorphic phantom exposed to systematically reducing CT dose derived from an X-ray beam at 140kVp and varying current from 160mA (full diagnostic dose) to 10mA (ultra low dose). Preliminary results show that for the purpose of CT attenuation correction, it is possible to successfully produce attenuation maps at ultra low dose with very low error (compared to full diagnostic dose) if used with the segmentation method presented.

Abstract-In this work we consider a probabilistic methodology that models the intensity distributions found in pure and partial volume (PV) voxels. We introduce some methodological developments that enable explicit modeling of the PV voxels prior Probability Density Function (PDF). This new formulation can be applied generically across different imaging modalities including PET and SPECT. In this paper, we establish for the first time, that the prior PDF of voxels that arise from the PV effect in volumetric data can be well described by a simple phenomenological law called Benford's Law, which significantly eases parameter estimation compared to other methods. Results from simulated data are presented, along with a preliminary PET phantom study utilizing registered CT data to determine the quality of the resulting probabilistic voxel classification scheme.

Abstract-In this work we consider a probabilistic methodology that models the intensity distributions found in pure and partial volume (PV) voxels. We introduce some methodological developments that enable explicit modeling of the PV voxels prior Probability Density Function (PDF). This new formulation can be applied generically across different imaging modalities including PET and SPECT. In this paper, we establish for the first time, that the prior PDF of voxels that arise from the PV effect in volumetric data can be well described by a simple phenomenological law called Benford's Law, which significantly eases parameter estimation compared to other methods. Results from simulated data are presented, along with a preliminary PET phantom study utilizing registered CT data to determine the quality of the resulting probabilistic voxel classification scheme.