Motion compensated fully 3D list-mode reconstruction

Ziel/Aim:

PET investigations of the brain, e.g. in receptor studies, typically exhibit long acquisition times making them sensitive to patient motion during the data acquisition and leading to potentially significant motion blur of the reconstructed activity distribution. This circumstance makes motion compensation one of the most pressing problems in high resolution PET scanners. We report on our work concerning integration of event-based motion compensation algorithm into a previously developed high resolution fully 3D list-mode reconstruction.

Methodik/Methods:

We integrated motion compensation into our implementation of 3D list-mode Ordered Subsets Expectation Maximization reconstruction (3D LMOSEM). The system matrix elements are computed on-the-fly and are modelled to be proportional to the intersection volume of voxels with Lines Of Response having a finite cross section (TOR: Tube Of Response). Motion information with high temporal resolution is provided by an external infra-red motion tracking system and used by the motion compensation algorithm for suitable spatial transformation of the individual TORs. Scatter correction factors (calculated using Single Scatter Simulation) are reconstructed into a scatter image, which is afterwards used in the forward projection step of the reconstruction together with an image of the delayed events. We have evaluated the new method in phantom and patient studies by comparing it with the previously developed procedure of using standard sinogram-based OSEM-reconstruction after motion pre-correction of the list mode data. The evaluation procedure has been divided into two parts: i) quantitative accuracy and spatial resolution (FWHM comparison) in phantoms and ii) visual evaluation of patient brain studies.

Ergebnisse/Results:

The new method provides a significant improvement of the reconstructed resolution (more than 25%) in comparison to the standard reconstruction of our scanner (Siemens ECAT HR+) while maintaining a high level of quantitative accuracy. Visual evaluation of five F-DOPA and five FDG brain studies showed substantially higher level of details in several brain structures as well as almost complete elimination of motion related artefacts.

Schlussfolgerungen/Conclusions:

The developed reconstruction allows almost complete elimination of motion blurring artefacts in brain studies and provides substantially better resolution than the standard reconstruction of the Siemens HR+ scanner. However, high computational complexity of the algorithm leads to substantial reconstruction times, which currently prevent its use for on-the-fly reconstruction in clinical routine. Further developments are necessary to eliminate this