TY - JOUR T1 - Improved 3D statistical reconstruction of high resolution PET/CT images using anatomical priors and CT-based attenuation correction JF - Journal of Nuclear Medicine JO - J Nucl Med SP - 95P LP - 95P VL - 48 IS - supplement 2 AU - Juan Ortuño AU - George Kontaxakis AU - José Rubio AU - Pedro Guerra AU - Andrés Santos Y1 - 2007/05/01 UR - http://jnm.snmjournals.org/content/48/supplement_2/95P.1.abstract N2 - 320 Objectives: PET/CT high resolution tomographs provide anatomical information of small rodents, which is co-registered with functional dynamic PET data. We propose a novel algorithm which upgrade the resolution-noise ratio of fully-3D OSEM reconstructions of PET acquisitions, using additional data provided by the CT scan to minimize the cross-entropy of a weighted sum of the anatomical and functional images and correct the attenuation effect. Methods: The realistic MOBY mouse phantom has been simulated under a four-head rotating PET scanner with LSO pixelated crystals with a size of [ 1.5, 1.1, 12] mm3 and a useful FOV of [45,45,45] mm3, using the SimSET package, which includes scatter, attenuation and coincidence non-collinearity effects. The CT image was assumed to be the MOBY attenuation map at 44.7 keV. Nearly 7 million events were collected and binned into 4D sinograms. The system matrix was modeled with Montecarlo methods including crystal attenuation effects, and stored in sparse matrix format to get short reconstructions times. The proposed reconstruction method is a 3D and ordered subsets version of the minimum cross-entropy algorithm (OS-MXE) where the prior image model is weighted with a non-linear edge preserving operator over the anatomical image, and the system matrix values are also modified according to the attenuation coefficients obtained through the segmented CT scan. Results: The obtained images were compared against FORE+2D-OSEM, SSRB+2D-OSEM and 3D-MAP-OSEM reconstructions without anatomical information. In all cases the new method increases significantly the signal-noise ratio, preserving at the same time the tissue boundaries when they are associated to different emission rates, assuming that the alignment between the PET and CT scan is less or equal than intrinsic PET resolution. In addition, the algorithm does not include relevant artifacts in those tissue gaps without differential PET activity. Conclusions: The performed simulations suggest that a registered CT scan, if available, can improve significantly the quality of PET statistical 3D reconstructions in terms of noise reduction and tissue uniformity. ER -