The present study focused on SUVmax in patients with CRLM, examining the influence of PSF and TOF reconstruction algorithms as well as different TBR.
The integration of PSF and TOF into reconstruction revealed relevant impact on SUVmax when lesions were separated by their TBR (low, <4.8; high, >4.8) measured in 3D-OSEM reconstructed data. In lesions with a high TBR, both PSF + TOF and PSF showed significantly higher SUVmax compared to the corresponding non-PSF algorithms. The median relative differences were slightly higher for combined PSF + TOF (PSF + TOF vs. 3D-OSEM + TOF, +9.1%) than for PSF alone (PSF vs. 3D-OSEM, +6.4%). Of course, these results are, strictly speaking, only valid for the specific scanner and reconstruction software used in this study, but it can be expected that with other systems, similar SUV deviations would occur.
Knäusl et al. evaluated nine lung lesions regarding SUVmax based on PSF (also Siemens TrueX® algorithm) and OSEM reconstruction and reported even higher differences of 19% on average [12]. This may be due to typically higher TBR in pulmonary lesions with pronounced SUV differences and due to an increased noise level caused by the higher product of iterations and subsets used by Knäusl et al. (4 iterations, 21 subsets). Akamatsu et al. analyzed 41 lymph node metastases and observed SUVmax differences of about +40% between PSF + TOF and OSEM + TOF as well as between PSF and OSEM. Again, differences increased with an increasing number of iterations [16].
However, the observed inter-method differences related to PSF were significantly correlated with the lesions’ TBR and were significantly lower in lesions with a low TBR. A recent study on FDG-PET phantom measurements analyzed radial activity concentration profiles of spheres filled with a solution of F18-FDG at three different SBR. It demonstrated that signal elevations at the spheres’ boundaries (known as Gibbs artifacts [17]) occurred only in PSF + TOF and PSF and only at medium (6:1) and high (16:1) SBR. In analogy to the current results, this led to an artificially increased SBR and to the observed deviations in quantitative parameters [13].
The integration of TOF analysis resulted in SUVmax differences comparable to those observed for additional PSF (PSF + TOF vs. PSF, +10.4%; 3D-OSEM + TOF vs. 3D-OSEM, +8.6%). However, these deviations were measured in lesions with a low TBR while SUV in high-contrast lesions were significantly less affected by TOF. These inverse findings of PSF and TOF may explain why no correlation was observed for differences between combined PSF + TOF and 3D-OSEM (non-PSF, non-TOF) and the TBR. In other words, PSF + TOF increased SUVmax across the entire range of the TBR when compared to 3D-OSEM (correlation plot not shown). As a consequence, a differentiated assessment of PSF + TOF and PSF with regard to varying TBR is required. There are scarce data on the independent influence of TOF analysis on SUV in clinical lesions. In the abovementioned study, Akamatsu et al. reported SUVmax differences of only +2% between PSF + TOF and PSF as well as between OSEM + TOF and OSEM, probably due to relatively high TBR of the analyzed lesions. Nevertheless, a TOF-related SUVmax increase was mainly observed in lesions with low SUVmax [16]. Schaefferkoetter et al. assessed the signal-to-noise ratios (SNR) of FDG-avid foci which were artificially added to FDG-PET scans of 23 patients. The authors reported increasing SNR of OSEM + TOF, PSF, and PSF + TOF compared to OSEM. In analogy to the current results, the relative SNR gain of OSEM + TOF vs. OSEM was higher in lesions with lower count rate (20,000 vs. 60,000 counts) [18]. Taniguchi et al. performed phantom measurements (lesion diameter, 10 to 37 mm) and clinical studies in hepatic lesions (average diameter, 10.7 mm) to analyze the influence of PSF and TOF on lesion contrast, coefficient of variation (CV) of the background activity, as well as lesion SNR. Both PSF and TOF independently increased liver lesion SNR with the best tradeoff between lesion contrast and CV for combined PSF + TOF. Furthermore, PSF + TOF reduced the CV of the liver in overweight patients (>70 kg) to the CV level in OSEM reconstructed data of normal-weight patients (<70 kg) [19]. That such differences in quantitative measures can also result in improved lesion detection was shown by El Fakhri et al. who analyzed the influence of TOF integration on detection rates in simulated liver and lung lesions. TOF improved lesion detection over non-TOF PET in both hepatic and pulmonary lesions but showed the greatest advantage at low contrast (contrast, 2.0:1 vs. 5.7:1) and in patients with higher body mass index (BMI; >30 vs. <30) [20].
It is well known that the TBR is affected by the lesion size due to partial volume effects which are more pronounced in smaller lesions [21,22]. As we observed such a positive correlation between the lesion volume and the TBR, we included the TBR, the lesion volume, and their interaction into GLMs. For PSF integration, the GLM showed no association between SUVmax differences and the lesion size whereas a highly significant association with the TBR was observed. In addition, there was a significant interaction between the TBR and the lesion volume, indicating that the effect of the TBR on PSF-related SUVmax differences depends on the lesion volume. Thus, we observed an increased impact of the TBR in smaller lesions. Also, in TOF integration, the SUVmax differences showed a significant association with the TBR (PSF + TOF vs. PSF) or a tendency towards an association (3D-OSEM + TOF vs. 3D-OSEM). In comparison to PSF, a similar effect of the lesion volume can be observed in the interaction plots. However, in GLM, the interaction term showed no significance which may be caused by a high variance of SUVmax differences in combination with a small sample size. This dependency of TOF-related effects on the lesion volume is in agreement with previous studies showing that the impact of TOF is especially relevant in small lesions [23,24].
SUVmax are commonly used for threshold-dependent volume definition in a clinical setting. If the delineation is strictly based on SUVmax (i.e., relative threshold without background correction), these differences would also implicate corresponding MTV deviations as reported previously [12]. However, we refrained from volumetric analyses as such delineation methods may not reflect the clinical practice where more sophisticated algorithms with background correction or manual MTV delineation are required - especially in hepatic lesions [14,25,26]. Thus, the actual MTV deviations may be lower and less dependent on the TBR than the current results on SUVmax deviations suggest.
Nevertheless, these results underline that quantitative analyses in radiotherapy planning, follow-up, or multicenter studies can be distorted not only by different reconstruction settings but also when comparing lesions located in different organs (e.g., lung and liver) or one lesion with varying TBR measured over the course of time. Depending on the lesions’ TBR, one must be aware of SUVmax deviations mainly caused by PSF or TOF, respectively. This may be particularly true for hepatic lesions that feature a range of TBR in which both PSF- and TOF-related effects are relevant.
The present study is limited by the retrospective inclusion of only 15 patients with 28 lesions which may impair an accurate interpretation of the data that were characterized by relatively large IQRs.