Investigating the limits of PET/CT imaging at very low true count rates and high random fractions in ion‐beam therapy monitoring

Abstract

Under the poor statistical conditions in PET-based treatment monitoring, improved results can be achieved by considering PSF and TOF information during image reconstruction and by applying less iterations than in conventional nuclear medicine imaging. Geometrical fidelity and image noise are mainly limited by the low number of true coincidences, not the high LSO-related random background. The retrieved results might also impact other emerging PET applications at low counting statistics.

Initial clinical evaluation of PET‐based ion beam therapy monitoring under consideration of organ motion

“…In our previous work, these settings have been investigated in detail and found optimal under the given statistical conditions of small true count numbers and high random fractions. 28 The latter ones were found to range from about 70%-80% in this study due to the constant random background emerging from the intrinsic activity of the scanner lutetium oxyorthosilicate (LSO) crystals. The voxel size was set to 4×4×3 mm 3 in order to maximize the number of counts per voxel.…”

“…Remaining differences are mainly attributed to inaccuracies in the biological washout model 2,9,35 and the image noise at the encountered low counting statistics. 28 In terms of PET-based range assessment, a small, clinically not relevant (i.e., covered by the applied safety margins in beam direction) mean deviation of 1.6±4.6 mm was found between measurement and simulation. Still, a region of 5-20 mm over-ranges in the PET measurement exists due to slight deviations in the outer patient contour on the treatment day, as confirmed by the proximal CT profile comparison (Fig.…”

“…Such discrepancies clearly exceed the uncertainties in the simulated amount of β + -emitter and the activity quantification by the PET/CT scanner, which altogether are expected below 10% on the basis of dedicated phantom irradiation and imaging studies. 23,28 Consequently, the observed deviations are mainly attributed to inaccuracies in the washout modeling. In the mobile PTV region, activity quantification accuracy might be further degraded due to motion blurring and, in the case of motion-compensated reconstruction, changes in the correlation of internal and external motion during the 30 min PET data acquisition.…”

4D ML reconstruction as a tool for volumetric PET‐based treatment verification in ion beam radiotherapy

“…Random noise in each bin of Measured and Expected PET sinograms was obtained from a lognormal distribution whose parameters were estimated on the background of a set of posttreatment PET acquisitions performed at HIT (Heidelberg Ion beam Therapy Center, Germany). 13 It is worth noticing the need to add LSO background noise also to Expected PET sinograms. This imposes a coherence between Measured PET and Expected PET frames in the nonsignal area and makes the cost function L mainly depending on the inconsistencies in the signal area.…”

“…Kurz et al showed that the bias can be partially reduced by introducing point spread function (PSF) and time of flight (TOF) information in the reconstruction. 13 An alternative idea to face the low counting statistics was introduced by Gianoli et al: 9 the Expected PET can be used not only as the reference to compare Measured PET with, but also as "a guide" to help Measured PET reconstruction. Furthermore, by estimating the sinogram of the Expected PET, 14 differences between Expected and Measured PETs can be exploited not only in image but also in sinogram domain, thus allowing to act on the reconstruction process.…”