The role of activity, scan duration and patient's weight in the optimization of 18FDG imaging protocols on a TOF-PET/CT scanner

Background : Time-of-flight (TOF) PET technology determines a reduction in the noise and improves the reconstructed image quality , in low counts acquisitions, such as in overweight patients, allowing a reduction of administered activity and/or imaging time. However, international guidelines and recommendations on 18 F-fluoro-2-deoxyglucose (FDG) activity administration scheme are old or only partially account for TOF technology and advanced reconstruction modalities. The aim of this study was to optimize FDG whole-body studies on a TOF PET/CT scanner by using a multivariate approach to quantify how physical figures of merit related to image quality change with acquisition/reconstruction/patient-dependent parameters in a phantom experiment. Methods : The NEMA-IEC body phantom was used to evaluate contrast recovery coefficient (CRC), background variability (BV) and contrast-to-noise ratio (CNR) as a function of changing emission scan duration (ESD), activity concentration (AC), target internal diameter (ID), target-background activity ratio (TBR), and weight. The phantom was filled with 5.3 kBq/mL of FDG solution and the spheres with TBR of 21, 9, and 5 in 3 different sessions. Images were acquired at varying activity concentration from 5.1 to 1.3 kBq/mL and images were reconstructed for ESD of 30-151 seconds per bed position with and without Point Spread Function (PSF) correction. The parameters were all considered in simultaneous experiments and in a single analysis using multiple linear regression methods. Results : As expected, CRC depended only on sphere ID and on PSF application, while BV depended on sphere ID, ESD, AC and weight of the patient, in order of decreasing relevance. Noteworthy, ESD and AC resulted as the most significant predictors of CNR variability with a similar relevance, followed by the weight of the patient and TBR of the lesion. Conclusions Due to the interchangeable role of AC and ESD in modulating CRC, ESD could be increased rather than AC to improve image quality in overweight/obese patients to fulfil ALARA principles.

Background : Time-of-flight (TOF) PET technology determines a reduction in the noise and improves the reconstructed image quality , in low counts acquisitions, such as in overweight patients, allowing a reduction of administered activity and/or imaging time. However, international guidelines and recommendations on 18 F-fluoro-2-deoxyglucose (FDG) activity administration scheme are old or only partially account for TOF technology and advanced reconstruction modalities.
The aim of this study was to optimize FDG whole-body studies on a TOF PET/CT scanner by using a multivariate approach to quantify how physical figures of merit related to image quality change with acquisition/reconstruction/patient-dependent parameters in a phantom experiment.
Methods : The NEMA-IEC body phantom was used to evaluate contrast recovery coefficient (CRC), background variability (BV) and contrast-to-noise ratio (CNR) as a function of changing emission scan duration (ESD), activity concentration (AC), target internal diameter (ID), target-background activity ratio (TBR), and weight. The phantom was filled with 5.3 kBq/mL of FDG solution and the spheres with TBR of 21, 9, and 5 in 3 different sessions. Images were acquired at varying activity concentration from 5.1 to 1.3 kBq/mL and images were reconstructed for ESD of 30-151 seconds per bed position with and without Point Spread Function (PSF) correction. The parameters were all considered in simultaneous experiments and in a single analysis using multiple linear regression methods.
Results : As expected, CRC depended only on sphere ID and on PSF application, while BV depended on sphere ID, ESD, AC and weight of the patient, in order of decreasing relevance. Noteworthy, ESD and AC resulted as the most significant predictors of CNR variability with a similar relevance, followed by the weight of the patient and TBR of the lesion.
Conclusions : Due to the interchangeable role of AC and ESD in modulating CRC, ESD could be increased rather than AC to improve image quality in overweight/obese patients to fulfil ALARA principles.

Background
Thanks to the improvements in hardware components and in imaging reconstruction techniques, The main motivation for TOF-PET has always been the potential image quality improvement or reduction in image acquisition time [1,2]. The effective sensitivity gain was already described nearly 40 years ago [8,] as depending on the ratio between the object size D and the spatial FWHM of the TOF kernel Δx.
In oncology practice, typically a longer acquisition time is needed for a larger patient characterized by higher attenuation. Often the longer acquisition time does not compensate for the poor quality of the data. Because of the higher attenuation, larger patients are affected by more noise. TOF acts as an equalizer, bringing the image quality in larger patients closer to that in patients of average size [,].
Thus, the first consequence of TOF technology is that SNR gain is increased, and this is especially more evident for larger patients [2, ].
This improvement has been used in the clinical setting predominantly to reduce the imaging time [1,2]. However, international guidelines and recommendations on FDG activity administration scheme are rather old [] or only partially account for time-of-flight technology [] and advanced reconstruction modalities. Therefore, precise information on how to tune administered FDG activity and emission scan duration in whole-body oncological studies on TOF-PET/CT scanners, is still a demanding need for the nuclear medicine physicians.
Few papers in the literature studied the optimization of 18 FDG activity administration [,] or emission scan duration [] on TOF-PET/CT scanners, but they examined the two factors independently.
The aim of this work was to describe how the physical figures of merit related to PET image quality change with different acquisition, reconstruction and object dependent parameters on a TOF-PET/CT scanner. The study was designed to simultaneously analyse the impact of the different factors with a multivariable approach, using phantoms with a variable weight, which hosted several well-defined target sizes with a known target-to -background ratio, as done in previous study []. We selected the emission scan duration (ESD), the FDG activity concentration (AC), the target-to-background activity concentration ratio (TBR), the target size (ID), the weight (W) of scanned object and the application of

Phantom setup
The NEMA International Electrotechnical Commission (IEC) Body Phantom with 18 F solution was used.
The IEC phantom has an interior cavity volume of 9947 mL and contains 6 fillable spheres with 10, 13, 17, 22, 28, and 37 mm inner diameters (ID). A cylindrical insert filled with low density foam (density of 0.30 g/cm 3 ) was fixed along the centre of the phantom. Four micro-hollow spheres with ID of 4.1, 4.7, 6.5, and 8.1 mm were fixed to a foam support attached to the lung insert at the bottom of the phantom. The IEC phantom was centred in the transverse FOV of the scanner with the equatorial plane of the standard spheres coplanar to the centre of the axial FOV.
To simulate the activity outside the scanner FOV, the scatter phantom (Data Spectrum Corporation) was placed close at the end of the IEC phantom. It is a solid circular cylinder composed of polyethylene with outside diameter of 203 mm and a length of 700 mm. A 6.4 mm hole is drilled along central axis of the cylinder. A 700 mm polyethylene tube with an inside diameter of 3.2 mm and an outside diameter of 4.8 mm is placed in the hole.
Finally, to simulate a different patient habitus (W), a belt of 11 water bags of 500 mL and 3 cm thick was fit over the IEC phantom, resulting in an additional weight of 5.5 kg. Thus, while the IEC phantom simulates the standard 70 kg weight man, the IEC fitted with the additional belt represents a 108.5 kg weight man.

Phantoms preparation and acquisition
The spheres of the IEC phantom were filled with 18

Image reconstruction
After correction for attenuation, scatter, random, detector normalization, isotope decay, system dead time and crystal timing, images were reconstructed using a TOF, list-mode, blob-based, ordered  Table 1. The PSF, speed and smoothing filter parameters were kept fixed for each reconstruction, as we demonstrated previously that no significant difference in contrast recovery coefficient and background variability exists by changing speed, smooth and PSF values [19]. Table 1 Reconstruction parameters values used in phantom image reconstruction Box and whiskers plots were used to provide a univariate graphical representation CRC and BV with respect to significant predictors, identified by the regression models. Outliers and extremes are points higher than the value of the 75th percentile plus 1.5 or 3 times the interquartile distance, or lower than the value of the 25th percentile minus 1.5 or 3 times the interquartile distance, respectively.
The impact of the different acquisition and object dependent parameters on CNR, was further investigated by a multiple way principal effects ANOVA: acquisition and object dependent parameters were considered as independent variables (factors) and CNR as the dependent variables A post-hoc test (Scheffe´ F test) was performed to identify the main sources of variability. If a significant F value was found for one independent variable, then this was referred as a main effect. When a main effect was found, then the Scheffe´ test was performed to compare the dependent variable upon the levels of the factor 2 × 2, thus identifying the main sources of variability. These comparisons were represented by drawing the least squares means, which are the best linear estimates for the marginal means in the ANOVA design, together with the standard errors of the means (and thus the 95% confidence intervals). The statistical analysis was performed with the software STATISTICA 6.0 (Statsoft Inc, USA).

Contrast recovery coefficient
The recovery of 18  impact on CNR about one half the one of ESD and AC. Post-hoc Scheffè test showed a statistically significant increase in CNR for every contrast between adjacent levels of AC in the range explored (p < 0.001) (Fig. 3a). When considering 2.2 kBq/mL, which represents the activity concentration 60 minutes post injection of 3 MBq/kg of 18F-FDG, the CNR mean value increases of about 19% when moving to an activity concentration of 3.1 kBq/mL (which correspond to an injection scheme of

kBq/ml).
A similar behaviour was observed for all the ESD, W and TBR contrast tested. Post-hoc Scheffè test showed a statistically significant increase in CNR for every contrast between adjacent levels of ESD ( Fig. 3b), weight (Fig. 3c) and TBR (Fig. 3d) in the range explored (p < 0.001).

Discussion
Precise information on how to tune acquisition and reconstruction parameters as well as FDG activity to administer to patients in whole-body oncological studies on TOF-PET/CT scanners is required to balance the improvement in the image quality with the opportunity of reducing the radiation dose burden in particular to patients frequently exposed to several radiological examination during their follow up, as part of the optimization process required by the Euratom Directive 2016-59 [23].
The image quality and lesion detectability in 18 F-FDG PET imaging are limited by the low signal-tonoise ratio and by the relatively low spatial resolution, which results in a partial-volume effect affecting lesion visualization and quantitation [2].
This work was aimed to characterize the quality of PET images for a TOF-PET/CT scanner in a wide range of acquisition, reconstruction and object dependent parameters in settings like those encountered in clinical practice, by means of a phantom study. CRC, BV and CNR, which is closely related to lesion detectability, were the figures of merit used to describe PET image quality. Our study used a multivariate approach to quantify how these figures of merit change as a function of ESD and AC for different target size, TBR, weight and under the effect of the point spread function modelling correction.
The main result of this study is that the CNR of FDG lesions depends on ESD and AC in a similar way (β = 0.53 and β = 0.51). This result is rather new: a previous study performed on a non-TOF PET scanner [18] concluded that the main predictor of CNR was ESD (β = 0.60) and only with a half of the explanatory power (β = 0.27) came AC.
Moreover, our results show also a significant increase in CNR for each increasing step in AC or ESD in the range explored, i.e. from 1.3 to 5.0 kBq/ml and from 30 to 151 sec. The visual inspection of phantom images confirmed that the image quality, in terms of the noise level and contrast, can be improved by increasing the AC (Fig. 4a) or ESD (Fig. 4b). This finding agrees with the clinical results reported in the recent paper of Prieto [16]. The author observed a statistically significant difference in both the image noise and the overall image quality indexes of PET images obtained after 18FDG activity administration of 5.2 and 3.7 MBq/kg on the Siemens mCT TOF-PET/CT scanner.
The third predictor of CNR was the weight of the phantom (β=-0.38), indicating that on average the CNR decreased by 28% with increasing the weight of the scanned object by a factor of 1.55. It is known that the TOF technology allowing the reduction of the uncertainty on the annihilation event acts as a noise equalizer and brings an overall gain in signal-to noise ratio, being this effect more evident for larger object [1,10]. However, our result is quite new, at least to our knowledge, and shows that notwithstanding the TOF technology, there is still a dependence of CNR on the weight of the imaged object. This means that there is an additional way to improve the lesion detectability for larger patients, depending on their weight. This result suggests the definition of specific acquisition protocols for oncological whole-body studies tailored on patient's habitus, rather than using a fixed ESD of 60 seconds per bed position, as suggested by the manufacturer.
The last predictor of CNR was the TBR of the lesion (β = 0.26). This dependence may be explained by the partial volume effect, which reduces the apparent activity concentration in the lesion in the reconstructed image preventing the recovering of the true amount of activity for structures less than twice the reconstructed image resolution. The major gain in CNR was observed for low TBR (or low SUV) values, as when moving from TBR 5 to 9 the CNR increases of about 19%, while when moving from TBR 9 to 21 the CNR increase is only of about 9% (Fig. 3d). However, this is an intrinsic characteristic of the lesion itself and cannot be managed in the optimization process.
The fitted multiple regression model of CNR based on these premises, accounts for more than two thirds of CNR variance (adjusted R 2 = 0.76).
From these findings, one can derive that it is possible to opportunely tune ESD and AC on patient's weight in order to keep constant the CNR, or the detectability level, on the PET images of this scanner.
As a typical example, let us consider the following situation: a lesion with a TBR of 9 (SUV = 8.8) in a standard 70 kg body weight patient injected with 3 MBq/kg of 18 F-FDG imaged for 60 sec, 60 minutes post injection. According to Eq.
(3), we should expect a CNR of 5.2. To obtain a similar value in CNR for the same lesion uptake in an obese 108 kg body weight patient injected with 3 MBq/kg of 18F-FDG, the patient should be scanned for 120 s. In an analogous way, one could double the activity administration scheme (i.e. 6 MBq/kg) to obtain the same CNR. However, from a radiation protection point of view, it would be more advisable to increase the ESD than the AC for an improvement in lesion detectability.
Another result of this study is the dependence of CRC on sphere ID (β = 0.68) and PSF application (β = 0.23). This result confirms the dependence of CRC already reported by Zorz et al. [19], even with a slightly different analytical expression with respect to (2). This may be explained by observing that in [19] only four spheres (ID = 10,13,17 and 22 mm) were included in NEMA-IEC phantom analysis and CRC fitting was almost perfectly linear (R adj 2 =0.93). In our study, on the contrary, CRC dependence on ID was analysed in a wider sphere dimension range, where CRC values assume a sigmoidal trend with respect to ID (Fig. 1a), explaining also the relatively low R adj 2 of 0.51.
The third parameter related to image quality we investigated, the BV resulted to be dependent on the ROI dimension for which is defined (β=-0.62), emission scan duration (β=-0.39), activity concentration (β=-0.31) and weight (β = 0.24), in order of decreasing relevance, this model explaining 77% of variance in BV (R adj 2 of 0.77). The strong dependence of BV on sphere ID is not new and was also found by Zorz [19] and Brambilla [18], with similar weights. Even if BV cannot be considered a descriptor of noise, it is interesting to note the equivalent dependence of BV on ESD (β = -0.39) and AC (β = -0.31), in analogy with the finding that CNR depends on both these parameters with similar weights, reported above. The impact of the weight of the phantom on BV, actually reinforces the same finding on CNR.
A limitation of this study must be acknowledged: the results on CRC, BV and CNR found in the present study strictly apply to this PET/CT scanner and to FDG oncological examinations: extrapolation of these results to different TOF-PET/CT scanners and to different radionuclides should be tested in advance before application.