Impact and correction of the bladder uptake on ¹⁸F-FCH PET quantification: a simulation study using the XCAT2 phantom

Positron emission tomography (PET) is a non-invasive imaging technique that visualizes the distribution of different molecules in the body providing functional and molecular information for different tissues. PET is routinely used for staging and treatment response evaluation in oncology (Jeraj et al 2008, Ben-Haim et al 2009) with fluorodeoxyglucose (FDG) as the most common radiotracer of choice. Despite the fact that traditional evaluation is usually performed by visual inspection of the images, the current potential of PET relies on its capability to provide quantitative information (El Naqa et al 2007), usually provided by a semi-quantitative parameter known as Standard Uptake Value (SUV). This parameter provides relatively objective tumour characterization, reliable differential diagnosis, and earlier evaluation and monitoring of treatment response (Boellaard 2009, Adams et al 2010).

Regarding prostate cancer, with around 250 000 new cases each year in the United States and mortality around 12% (Siegel et al 2012), an efficient and early detection of the disease is critical both for successful patient care and for the cost-efficiency of the health care systems. Imaging plays a critical role in correct diagnosis and staging, since the tumour treatment must be selected in strict dependence on the clinical stage and risk profile (Jadvar 2009). In this context FDG PET has shown very poor sensitivity and specificity on prostate cancer mainly due to two critical factors: some prostatic tumours do not show an elevated glucose consumption; many infectious processes such as prostatic benign hyperplasia show an increased FDG uptake producing false positives (Salminen et al 2002, Jadvar 2011).

During the last decade the development of novel radiotracers provided PET a more relevant role in prostate imaging. The most widespread of these tracers are those based on choline labelled with different radioisotopes (Hara et al 1998, DeGrado et al 2001, Chen et al 2012). Choline is a compound that supports the synthesis of cell membranes and thus proliferation. In the case of prostate cancer, it has been strongly related with elevated levels of choline uptake and certain choline metabolites, which can be used as potential prognostic biomarkers for the management of prostate cancer patients (Awwad et al 2012). The Fluorocholine (FCH) molecule labelled with ¹⁸F showed the highest biological compatibility with choline leading to a behaviour in the body very similar to that of natural choline (Hara 2001). An example of carcinoma detection on ¹⁸F-FCH PET with bladder accumulation is shown on figure 1, where the tumour is pointed with an arrow. A well-known drawback of ¹⁸F-FCH and other ¹⁸F-based radiotracers is their variable urinary excretion with high accumulation in the bladder, which can affect detectability (Schoder and Larson 2004, Massaro et al 2012). Reported solutions in order to mitigate this effect are mainly aimed at early acquisitions or bladder voiding by urinary catheterization (Witney et al 2012), but these solutions are not ideal for several reasons. On the one hand, the accumulation of radiotracer in the bladder is a patient-dependent process so early acquisitions cannot guarantee avoiding bladder accumulation (Massaro et al 2012). On the other hand, catheterization for bladder voiding is an invasive process that has been identified as a potential source of infection (Lo et al 2014). Despite this serious shortcoming, ¹⁸F-FCH is currently the radiotracer of choice for prostate PET imaging with indications for diagnosis, staging, restaging, and therapy monitoring in prostate cancer (Hodolic 2011, Beheshti et al 2013, Chondrogiannis et al 2014). The usage of bladder voiding, early acquisitions or both is highly centre-dependent.

Zoom In Zoom Out Reset image size

Regarding quantification, ¹⁸F-FCH PET images have shown high SUV variability in normal prostate tissue (SUV 3.4–4.1), prostate cancer (SUV 1.7–6.2), and local recurrence (SUV 2.7–12.42) (Tindall and Scardino 2011). The reasons for this variability are still unclear, since SUV can be affected by different factors. Verwer et al (2015) modelled the ¹⁸F-FCH-PET kinetics of lymphatic and haematogenous metastases and proposed modifications in the SUV calculations to address problems related with ¹⁸F-FCH-PET kinetic studies. In this context, additional work is required for evaluating technical issues related to acquisition, reconstruction, and quantification (Boellaard 2009, Adams et al 2010, Silva-Rodríguez et al 2015). Particularly important in ¹⁸F-FCH-PET imaging might be the overestimation due to the inclusion of spill-in counts originally coming from separated high-activity regions such as the bladder. This effect has been defined in the literature as spill-in (Bai et al 2013) or shine-through effect (Liu 2012) and included in a broader definition of the partial volume effect (PVE) by different authors (Soret et al 2007). This effect has been widely reported in cardiac SPECT (DePuey et al 2012), dual tracer parathyroid scintigraphy (Wu et al 2003, Liu et al 2005), and ¹⁸F-FDG PET (Liu 2012, Du et al 2013). In PET, different cases have been reported to affect the quantification of lung tumours when the lesions are in the vicinity of the myocardium or the quantification of prostate tumours close to the urinary bladder (figure 1). Despite multiple methodologies that have been proposed for the correction of the underestimation produced by the spill-out counts in small regions (Erlandsson et al 2012), which responds to the most traditional definition of PVE (Rousset et al 1998), the effect of spill-in counts has been less studied and it is not routinely corrected for in PET studies.

The aim of this work is first to study the impact of the spill-in counts from the bladder on the quantitative evaluation of ¹⁸F-FCH PET studies using realistic anthropomorphic computational simulations, and second, to propose a correction method for this effect when necessary. To the best of our knowledge, further analyses on this effect or similar correction methods have not been previously reported in literature.

Both geometric and anthropomorphic physical phantoms have been widely used for investigating the impact of different factors on the output PET images, enabling for us applications that cannot be performed with patient studies. Nevertheless, they have several important limitations, such as reduced flexibility for changing shapes and volumes of the internal structures, high cost, and cumbersome use. An alternative is the use of digital phantoms (Zaidi and Xu 2009), so that simulated PET studies can be generated from the realistic projection of digital phantoms by using computer simulation techniques. In this work, we generated a database of ¹⁸F-FCH PET studies by Monte Carlo (MC) simulation of a modified version of the XCAT2 phantom (Extended Cardiac-Torso Phantom Version 2.0) (Segars et al 2010). Figure 2 shows a general layout of the experiment, which allowed us to investigate the impact of bladder uptake on prostate SUV values in a realistic and well-controlled framework.

Zoom In Zoom Out Reset image size

An anthropomorphic phantom for ¹⁸F-FCH PET imaging

Monte Carlo simulation

Reconstruction

Correction method for spill-in counts effect

For the current work, we have modified the original OSEM algorithm in equation (1), where $~\widehat{f}_{j}^{n}$ is the value of the reconstructed image at pixel j for the n iteration, ${{p}_{i}}$ is the measured projection at the i bin, and ${{H}_{ij}}$ is the detection probability of photons from pixel j to projection bin i, for including a novel practical method for bladder uptake correction. The estimation of ${{H}_{ij}}$ is based on the multiple ray-tracing techniques included in the STIR library and it was chosen in order to reproduce the GE Advance protocols. No dedicated Point-Spread-Function (PSF) was included for the particular scanner in the ${{H}_{ij}}$.

$$\begin{eqnarray}&&\widehat{f}_{j}^{n+1}=\frac{\widehat{f}_{j}^{n}}{\sum_{{{i}_{2}}\in {{S}_{b}}}{{H}_{{{i}_{2\,}}j}}}\underset{i\in {{S}_{b}}}{\mathop \sum}\,{{H}_{ij}}\frac{{{p}_{i}}}{\sum_{k}{{H}_{ik}}\widehat{f}_{k}^{n}}.\end{eqnarray} \tag{ 1 }$$

The method requires an independent estimate of the bladder volume, shape, and activity that is obtained from the reconstructed image. An operator drew an ROI R visually containing the activity in the bladder. The segmented bladder was then obtained as shown in equation (2), where ${{R}_{j}}$ is the ROI pixel value (0 or 1) for pixel j, $\widehat{f}_{j}^{N}$ the final reconstructed image value and ${{B}_{j}}$ the pixel value on the segmented bladder, $B$. Different operators manually delineated the ROIs for each simulation, so that the corrected image measurements include potential variability induced by manual delineation of the bladder. The segmented bladder, $B$, was then analytically forward-projected into the sinogram space following equation (3), where ${{P}_{i}}$ is the contribution of the bladder to bin i. This contribution is then included in the reconstruction algorithm as a background term as shown in equation (4). A final reconstruction is performed providing reconstructed images without the bladder contribution.

$$\begin{eqnarray}&&{{B}_{j}}={{R}_{j}}\widehat{f}_{j}^{N}\end{eqnarray} \tag{ 2 }$$

$$\begin{eqnarray}&&{{P}_{i}}=\underset{j}{\mathop \sum}\,{{H}_{ij}}{{B}_{j}}\end{eqnarray} \tag{ 3 }$$

$$\begin{eqnarray}&&\widehat{f}_{j}^{n+1}=\frac{\widehat{f}_{j}^{n}}{\sum_{{{i}_{2}}\in {{S}_{b}}}{{H}_{{{i}_{2}}\,j}}}\underset{i\in {{S}_{b}}}{\mathop \sum}\,{{H}_{ij}}\frac{{{p}_{i}}}{\sum_{k}{{H}_{ik}}\widehat{f}_{k}^{n}+{{P}_{i}}}.\end{eqnarray} \tag{ 4 }$$

This approach is inspired by the methodology to correct physiological background radioactivity on dynamic acquisitions originally proposed by Tsoumpas and Thielemans (2009). A schematic layout of the process can be observed on figure 3.

Zoom In Zoom Out Reset image size

Tumour quantification

Impact of bladder uptake

As an example, figure 4 shows the reconstructed images of the tumour with SUV 3.03 g l⁻¹ and different bladder uptakes ranging from 1.01 to 15.92 g l⁻¹ using a full range scale. As expected, an increase in bladder uptake is a counterpart for detectability, so that the tumour with SUV value of 3.03 g l⁻¹ is clearly observed for bladder uptake of 1.01 g l⁻¹ but not for high bladder uptakes.

Zoom In Zoom Out Reset image size

Figure 5 shows the measured tumour SUV_max and SUV_mean for different bladder uptakes ranging from 1.01 to 18.19 g l⁻¹ obtained from the simulated studies with theoretical tumour SUVs of 3.03 and 6.06 g l⁻¹. It is observed that measured tumour SUVs values were significantly increased with respect to the originally measured tumour SUV value when bladder SUV is increased. Biases of the values are shown as percentages on table 1. Measured SUV_mean values underestimated the theoretical SUV in all cases, due to the small tumour size. The results were compared to the SUV_mean without bladder uptake (simulated bladder uptake 1.01 g l⁻¹) in order to evaluate repeatability. The uncertainty bars in the figure were calculated for both SUV_max and SUV_mean as the standard deviation (STD) of the SUV measured for each of the five repetitions performed for each case. It can be derived from the results on figure 5 that this effect is particularly important when simulated bladder SUV is higher than theoretical tumour SUV, and that SUV_mean is less sensitive to this particular effect than SUV_max. The fact that the maximum bias in the same range of bladder values is around half for a lesion with two times the simulated SUV (23.4% versus 41.3% for SUV_max and 10.8% versus 22.2% for SUV_mean) is a suggestion of a strong dependency of the effect on the bladder/tumour ratio.

Table 1. Biases of the measured tumour SUV_max and SUV_mean for all bladder and tumour combinations.

Bladder SUV (g l−1)	Simulated tumour SUV (g l−1)	SUVmax bias (%)	SUVmean bias (%)
1.01	3.03	1.7	0
3.03	3.03	5.6	9.1
7.58	3.03	19.8	13.1
13.64	3.03	24.1	16.7
15.92	3.03	27.4	20.2
18.19	3.03	41.3	22.2
1.01	6.06	−2.3	0
3.03	6.06	0.1	−1.8
7.58	6.06	0.7	0.8
13.64	6.06	7.9	6
15.92	6.06	16.2	8.5
18.19	6.06	23.4	10.8

Zoom In Zoom Out Reset image size

Impact of bladder volume

Figure 6 shows variations of the measured SUV_max with bladder volumes ranging from 100 to 750 ml, for simulated tumour SUVs of 3.03 and 6.06 g l⁻¹ and bladder SUV of 7.58 g l⁻¹. Our results did not show any important variations with the bladder size for the analysed volumes ranges.

Zoom In Zoom Out Reset image size

Correction method

The proposed correction method was applied to all the simulated images and SUV measurements were performed again over the corrected images. Figure 7 shows an example of the visual effect of the correction and the improvement of detectability. The bladder has disappeared and the tumour is observed more clearly on the final image.

Zoom In Zoom Out Reset image size

Regarding quantification, figure 8 shows the effect of the correction on quantification for lesions SUVs of 3.03 and 6.06 g l⁻¹, showing plots of the values obtained with and without the correction. It can be observed that the error induced by the bladder has disappeared after correction. The biases induced by the bladder, shown in table 2, decreased significantly, presenting values fewer than 4% for SUV_mean and fewer than 6% for SUV_max. It is observed in the uncertainty bars that the correction did not decrease the repeatability of the measurement.

Table 2. Bias of the measured tumour SUV_max and SUV_mean for the bladder and tumour combinations including values for corrected and uncorrected images.

Bladder SUV (g l−1)	Simulated Tumour SUV (g l−1)	SUVmax bias (%)	Corrected SUVmax bias (%)	SUVmean bias (%)	Corrected SUVmean bias (%)
1.01	3.03	1.7	1.7	0	0
3.03	3.03	5.6	4.3	9.1	2.0
7.58	3.03	19.8	5.5	13.1	0.5
13.64	3.03	24.1	2.1	16.7	−2.0
15.92	3.03	27.4	0.5	20.2	−3.5
18.19	3.03	41.3	4.0	22.2	0.5
1.01	6.06	−2.3	−2.3	0	0
3.03	6.06	0.1	1.8	−1.8	−1.8
7.58	6.06	0.7	2.0	0–8	0.8
13.64	6.06	7.9	0.8	6.0	−2.8
15.92	6.06	16.2	1.7	8.5	−1.8
18.19	6.06	23.4	2.7	10.8	−0.2

Zoom In Zoom Out Reset image size

In order to validate the impact of the correction on the convergence of the reconstruction method, we selected the case of the tumour SUV of 3.03 g l⁻¹ and bladder SUV of 7.58 g l⁻¹ (figure 7). Figure 9 shows both the SUV_mean and $\Delta \text{SU}{{\text{V}}_{\text{mean}}}$ defined in equation (5), up to 64 sub-iterations.

$$\begin{eqnarray}&&\Delta \text{SU}{{\text{V}}_{\text{mean}}}\left(\%\right)=\frac{\text{SU}{{\text{V}}_{\text{mean}}}\left(i{{t}_{n}}\right)-\text{SU}{{\text{V}}_{\text{mean}}}\left(i{{t}_{n-1}}\right)}{\text{SU}{{\text{V}}_{\text{mean}}}\left(i{{t}_{n}}\right)}\times 100\end{eqnarray} \tag{ 5 }$$

$\Delta \text{SU}{{\text{V}}_{\text{mean}}}~$ has values of 1.44% and 1.56% for 16 sub-iterations, 0.26% and 0.34% for 32 sub-iterations, and 0.18% and 0.24% for 64 sub-iterations for uncorrected and corrected images respectively, showing that the convergence of the reconstruction is similar for both corrected and uncorrected images and that the chosen 32 sub-iterations are reasonable for ensuring the convergence even when applying the proposed correction method.

Zoom In Zoom Out Reset image size

Bladder accumulation is a well-known drawback for ¹⁸F labelled radiotracers in general. When using ¹⁸F -FCH, this problem is usually solved by performing very early acquisitions or by bladder voiding using a catheter (Witney et al 2012), but these solutions are not optimal. On the one hand, the accumulation of radiotracer in the bladder is a physiological process, so it is patient-dependent, and early acquisitions have proven to be ineffective. Massaro et al (2012) concluded that the earlier and/or dynamic acquisitions do not provide further relevant information to change the exam interpretation. On the other hand, the use of a catheter for bladder voiding is effective, but it has been recognized as a potential source of infection (Lo et al 2014), and it is an invasive and patient disturbing process, when imaging should be kept as a non-invasive technique whenever possible. In this investigation, we have developed and tested a novel software-based correction method that allows us to correct the effects of bladder accumulation on ¹⁸F-FCH PET, and that is potentially applicable to other situations. This methodology expands the scope of the work presented by Tsoumpas and Thielemans (2009).

The first step was to characterize the effect of bladder accumulation over the evaluation of tumours located in the region. The first consequence of this accumulation, the loss of detectability (Schoder and Larson 2004) on the bladder region is well known, observed in figure 4. Additionally, Liu (2012) reported that this accumulation could also have an impact on quantification values caused by shine-through effect or spill-in, presenting some clinical cases. In order to assess this effect on ¹⁸F-FCH studies, we designed a Monte Carlo simulation experiment, which produced images of the XCAT2 anthropomorphic phantom with a hot spot (24 ml) on the prostatic left lobe for different combinations of bladder accumulation intensities and volumes. The results of this analysis, shown in figure 5, revealed that the measured SUV values could be biased as much as 41.3% for SUV_max and 22.2% for SUV_mean when the bladder SUV varies on a range from 1.01 to 18.19 g l⁻¹, with strong dependence on bladder/lesion ratio. No remarkable variations of the measured SUV were found when the bladder volume ranges from 100 to 750 ml, as it is observed on figure 6, pointing out that the effect is independent of the size of the secondary source. These first results reveal that the quantification on the prostatic area is very sensitive to variations in bladder accumulation, which is dependent on technical and uncontrollable physiological factors, suggesting that uncorrected ¹⁸F-FCH might not be an appropriate radiotracer for quantification of tumours in this area, with ¹¹C-labelled choline analogues providing a more reliable solution. These conclusions provide supporting evidence of additional ¹⁸F-FCH quantification problems, recently reported by Verwer et al (2015). The obtained SUV values should be taken and used with precaution when any presence of radiotracer is observed in the bladder, especially if its intensity is higher than the intensity of the lesion. The prevention of this effect with bladder voiding or the usage of a proper correction method is mandatory. The results also showed a systematic underestimation of the SUV_mean values, most likely related to the partial volume effect (PVE) (Soret et al 2007) due to the small tumour size. This underestimation is common on clinical PET, but it is not a shortcoming for the use of SUV_mean in the clinic, where the most concerning issue is usually SUV repeatability (Silva-Rodríguez et al 2015).

After this in-depth analysis of the bladder uptake impact, we proposed a correction method. The method was based on the hypothesis that spill-in counts from the bladder could be corrected with appropriate modification of the reconstruction process. The bladder was segmented from the conventional reconstructed PET image using a manually drawn ROI. After the segmentation, the bladder was analytically projected with the same projector that was used for the iterative reconstruction, generating new analytical sinograms of the bladder only. Segmenting directly from a fused CT image would avoid the step of the first reconstruction, but since the bladder volume will vary during the acquisition, and as PET and CT acquisitions are not simultaneous, there will be a size mismatching between CT and PET, making the PET segmentation more convenient for the particular case of the bladder (Heiba et al 2009). This limitation may be diminished if simultaneous PET-MR systems are utilised, where the bladder could be accurately segmented from the MR co-registered image.

Following segmentation and forward projection, the reconstruction of the original sinograms was performed again, with the bladder sinograms as a physiological background term for the reconstruction (Tsoumpas and Thielemans 2009). This background term in the reconstruction algorithm provided the information that the bladder contribution of the sinogram is undesired on the final image. The process is mathematically the same as used for standard scatter and random coincidences corrections. This new method was applied to all the simulated images and the result was an image without the bladder contribution, as shown in figure 7. This first result is already useful, since it solves the problems of lesion detectability observed on figure 4 and produces images visually similar to those of ¹¹C-choline (Witney et al 2012). It is remarkable that the results are very consistent over all the images, even when manually drawn ROIs were used, highlighting that the method is not very dependent on ROI delineation, as soon as the ROI has been drawn in a reasonable way, taking almost all the activity of the bladder.

Despite these first promising results, the main goal of this work was to solve the problems with quantification. In order to evaluate this, we compared the SUV_max and SUV_mean quantification values before and after the correction. The results on figure 8 point out that the correction also manages to successfully recover the quantification values. A significant improvement was observed in SUV_max and SUV_mean for both lesions of SUVs of 3.03 and 6.06 g l⁻¹ with bladder/lesion ratios from 0.18 to 6. The main conclusion of the analysis is that after the correction, the images provided higher accuracy and repeatability of both SUV_max and SUV_mean values, with good repeatability and in good agreement with simulated SUV values and quantified values without bladder uptake respectively. About statistical noise, no increase was noticed compared to the images without the correction according to the STD over five repetitions. Convergence of the proposed methodology was studied and it was shown that the correction does not have an impact on the convergence of the reconstruction algorithm, so the chosen number of thirty-two iterations was adequate.

Overall, the results of our experiment show that the usage of the method makes possible a reliable use of SUV on prostatic lesions when injecting ¹⁸F-FCH, at least from a reconstruction point of view. More experiments are needed to ensure the applicability of the proposed method to patient studies, and modifications of the method might be necessary when applying it to the clinic, particularly related with the segmentation of the bladder that can be hindered by the physiology of clinical PET studies. These and other factors affecting the segmentation such as influence of patient movement and bladder volume variations are outside the scope of this work but subject to future investigations.

Finally, it has to be mentioned that the proposed methodology could be applied to other cases, such as spill-in counts derived from the myocardium to coronary arteries or a neighbouring lung tumour (Liu 2012). Nevertheless, potential motion artefacts could complicate the applicability of the method.

The impact of the bladder uptake on prostate tumours quantification was evaluated using MC simulations. Our results showed that measured SUV can be biased as much as 40% when bladder/tumour ratios are high. A correction method based on the introduction of prior information about bladder on the reconstruction was proposed and tested, showing improvements both on visual detectability and quantification. Applying the proposed correction method, reliable SUVs can be obtained from ¹⁸F-FCH images.

This work was supported in part by the public Fondo de Investigaciones Sanitarias (PI11/01806), MINECO (2014/CSUN1/000004) and the EU COST Action TD1007 (www.pet-mri.eu). Instituto de Salud Carlos III supports J. Pardo-Montero through a Miguel Servet grant (CP12/03162) and P. Aguiar is awarded a public fellowship from Xunta de Galicia (POS-A/2013/00).