Euromedim 2006 :1st European Conference on Molecular Imaging Technology

Image reconstruction techniques evaluation and validation for high resolution human brain PET imaging

11 May 2006, 09:30

15m

Palais du Pharo, Marseille

oral S5-S6 medecine Clinical Imaging

Speaker

Dr Claude Comtat (Frédéric Joliot Hospital Facility, CEA/DSV/DRM, Orsay, France)

Description

High resolution PET imaging is now a well established technique not only for small animal, but also for human brain studies. The ECAT HRRT brain PET scanner (Siemens Molecular Imaging) is characterized by an effective isotropic spatial resolution of 2.5 mm, about a factor of 2 better than for state-of-the-art whole-body clinical PET scanners. Although the absolute sensitivity of the HRRT (6.5 %) for a point source in the center of the field-of-view is increased relative to whole-body scanners (typically 4.5 %) thanks to a larger co-polar aperture, the sensitivity in terms of volumetric resolution (75 µm3 at best for whole-body scanners and 16 µm3 for the HRRT) is much lower. This constraint has an impact on the performance of image reconstruction techniques, in particular for dynamic studies. Standard reconstruction methods used with clinical whole-body PET scanners are not optimal for this application. Specific methods had to be developed, based on fully 3D iterative techniques. In this study, we present the performance of several implementations of the fully 3D sinogram-based OSEM (Ordered Subset Expectation Maximization) image reconstruction algorithm, with different noise (weighting scheme) and spatial resolution models. Their performance was evaluated for the neuronal dopamine transporter (DAT) imaging in humans with the HRRT, using a selective DAT radioligand ([11C]-PE2I). Eleven healthy volunteers were scanned, and the striatum binding potential (BP) of [11C]-PE2I was calculated using a simplified reference model with the cerebellum as the non-specific binding region. Another set of eleven healthy volunteers was scanned on the whole-body ECAT EXACT HR+ scanner (Siemens Molecular Imaging) using the same protocol. The BP values obtained on the HR+ with a well validated analytic reconstruction technique (3DRP) were considered as the reference. The BP values obtained with the HRRT were validated against the HR+ reference values by post-processing the HRRT images to match the partial volume effect of the HR+ images. The use of a simplified noise model in the HRRT reconstruction algorithm introduces a systematic quantitative bias in the activity measured in the image when the relative amount of random coincidences is high, for example at the beginning of a dynamic acquisition. The modeling of a realistic intrinsic spatial resolution allows noise amplification in the image to be reduced, in particular at very low count levels (for frame duration of the order of one minute or later frames for 11C labeled radioligands). These results validate the use of the HRRT for cerebral DAT imaging in humans, and show the importance of using a realistic modeling of the acquisition process in the reconstruction algorithm for high resolution dynamic brain PET imaging.