Abstract
245
Objectives: We present results of tomographic reconstructions using point ensembles with application to computer simulated, 3D list-mode TOF-PET data.
Methods: The point-ensemble image is defined in terms of a set of spatial origins (points) of all detected events. This image consists of 3N parameters (3 spatial coordinates for every event, where N is the number of events). All possible locations of points constitute the point ensemble. A 3N-dimensional probability density function (PDF) is defined for the ensemble. Every state (a point in the 3N-dimensional space) of the ensemble is characterized by a probability which depends on the point locations, the scanner geometry, and scan data. The Markov Chain Monte Carlo algorithm is used to draw samples from the ensemble. A computer simulation of the acquisition of 10,000,000 events of a 3D list-mode TOF-PET scan was used. A mathematical phantom consisting of six superimposed ellipsoids and an ideal scanner with perfect efficiency (44-cm-radius ring with 16 cm axial span) was assumed. The accuracy and reconstruction times were estimated for several simulated values of TOF resolution, Δt.
Results: The method produces artifact-free images. The bias of the activity estimates within the hot spheres decreased with improved timing resolution of the TOF acquisition. The reconstruction times using an 2GHz Xeon processor varied from 20 to 90 minutes, depending on Δt.
Conclusions: Our approach allows straightforward and efficient implementation of complex acquisition geometries. The point-ensemble image representation provides a useful alternative to voxel-based image representations for the list-mode TOF-PET. An objective assessment of the point-ensemble image quality needs to be performed.
Research Support: NIH R21 CA123057
- Society of Nuclear Medicine, Inc.