Speaker
Mr
Cedric Lemaître
(Vrije Universiteit Brussel)
Description
The absence of very small crystal pixels in monolithic scintillation detectors has a number of potential
advantages such as higher sensitivity, better energy resolution and continuous coordinates. In such detectors,
the incidence position of the 511 keV photons on the detector surface is derived from the measured scintillation
light distribution. To extract this information, we used artificial neural networks
To this end, each detector module has to be position-calibrated by training the neural networks. When a neural
network is trained for a specific incidence angle, it yields immediately a DOI corrected incidence position of the
impinging photon.
An automated procedure to simultaneously obtain the calibration data to train all the neural networks for all
detector modules in a fully assembled PET system has been developed and evaluated on a simulator set-up.
After calibration, images of a point sources at various radial distances were taken to evaluate the quality of the
procedure.
Authors
Mr
Cedric Lemaître
(Vrije Universiteit Brussel)
Dr
Peter Bruyndonckx
(Vrije Universiteit Brussel)
Co-authors
Mr
D.J. (Jan) Van der Laan
(Technische Universiteit Delft)
Dr
Dennis Schaart
(Technische Universiteit Delft)
Mrs
Magalie Krieguer
(Vrije Universiteit Brussel)
Mr
Marnix Maas
(Technische Universiteit Delft)
Dr
Olivier Devroede
(Vrije Universiteit Brussel)
Prof.
Stefaan Tavernier
(Vrije Universiteit Brussel)