Floris Jansen^{1}, Mark Fries^{1}, Tuoyu Cao^{1}, Mehdi Khalighi^{2}, and Chang Kim^{1}

Accurate quantitation in PET requires good stability of the detector gain. The challenging thermal environment of the detector in a PET/MR system (proximity to gradients, induced eddy currents, heat from RF shield, ...) makes accurate temperature compensation important. Current solutions rely on characterization of detector response together with real time temperature measurement for a predictive (open loop) gain control. This work presents a method of gain control that operates in real time by analyzing spectral information of singles events, permitting closed loop gain control in the presence of temperature gradients or count rate variations.

[1] Kim et al., Compensation for thermally-induced loads on PET detectors from MR stimulus in simultaneous PET/MR, ISMRM 2014, p780.

[2] Levin et al., Design Features and Mutual Compatibility Studies of the Time-of-Flight PET Capable GE SIGNA PET/MR System. IEEE TMI. 2016; 35(8):1907-14

Figure
1: Photopeak (blue) and scatter (green) events pass the discriminator window. If the photopeak shifts by some amount Δ, additional
scatter (orange) will be admitted with lower energy and therefore different
distribution (shape) than expected by the scatter model. This can result in a small quantitation error.

Figure
2: the weighted sum of counts in windows performs a random walk – a
statistically significant shift occurs when the absolute value of the sum
changes more rapidly than the envelope of these curves would imply; this can be modeled with a $$$\sqrt{N}$$$ curve.

Figure
3: global peak drift during a thermal gradient. Top curve: sub-optimal
thermal compensation shows significant peak
drift. Bottom curve: using real time compensation, peak quickly stabilizes, then
remains very close to target value

Figure
4a-b: map of peak shifts for each device around detector ring (15 devices in z, 112 in x)
4a:
correction per block (block = 3
rows, 1 column): shifts showing residual structure
4b: correction per device (one
pixel in this image): no residual structure
4c: box plot of data in 4a confirms
systematic structure due to local thermal gradient
4d: blox plot of data in 4b confirms systematic
structure has been removed

Figure
5: example of singles spectrum showing 307 keV peak (Lu-176) is well resolved

Figure 6: Simulation of gain correction using
intrinsic radiation only. After initializing blocks with random gain errors
from -10% to +10%, algorithm tried to recover correct gain. Every dot in 6a
corresponds to a single correction being applied, and it shows that after just
a few seconds, the gain distribution has reduced significantly. Figure 6b:
after 10 seconds, the correction calculated is very well correlated to the
initial error applied – with residual RMS error 0.2%.