Keywords dose optimization - BGO-based positron-emission tomography - dependency - reconstruction
algorithm
Background
Positron emission tomography/computed tomography (PET/CT) corroborates anatomical
details by providing functional information. With the increasing number of clinical
applications of this imaging modality in oncology, patients require PET/CT scans more
frequently at various stages, such as initial staging, interim response, response
to therapy, and follow-ups.[1 ]
[2 ]
[3 ]
[4 ] Oncological imaging commonly utilizes an effective radiopharmaceutical, namely18 F-fluorodeoxyglucose (18 F-FDG).[5 ]
18 F-FDG is a fluorine radioisotope produced by cyclotrons that can be scanned 50 to
75 minutes after injection into a patient's body.[6 ] The management of the injected dose of 18 F-FDG for whole-body PET-CT scans follows the European Association of Nuclear Medicine
(EANM) guidelines.[6 ] The ENAM guidelines ensure that the measured FDG tumor uptake is within specific
limits (370–740 MBq), regardless of the type of device used or study location.[6 ]
[7 ] The new version of ENAM guidelines (2015) provides an overview of the earlier findings
and attempts to address some new developments in PET scans, such as time-of-flight
technology.[6 ] Few studies have aimed to optimize FDG examinations after the last update of the
EANM recommendations. According to a study, PET/CT with a bismuth germanium oxide
(BGO) detector can reduce the 18 F-FDG injection dose by up to 25% in patients with Hodgkin's lymphoma without sacrificing
image quality.[8 ] Nevertheless, this study focused on a single indication to optimize the administered
dose. Another study that used a four-ring lutetium-yttrium-orthosilicate (LYSO) TrueV
scanner (Siemens Medical Solutions, Knoxville, TN, United States) assumed that a modest
reduction of either injected FDG dose or the time per bed position to levels below
the limits provided in the EANM procedure guideline might be possible.[9 ] In the Fred Wickham study, another Siemens scanner (Biograph mCT Flow) was used
to establish an expression in terms of sex, height, and weight to optimize the injected
dose and acquisition times.[10 ] Reconstruction algorithms have been developed over the years to reduce errors and
artifacts and improve image quality. Ordered subset expectation maximization (OSEM)
is the most widely used algorithm for PET/CT scanning. Advances in OSEM and resolution
recovery methods, such as point spread function modeling, have improved PET image
quality by considering all statistical and physical processes during data acquisition.[11 ]
[12 ] PET-CT scanners have comparable image quality results depending on the technology
used for detecting tumors, in addition to data acquisition and reconstruction methods.
The acquisition time and injection dose are influenced by scanner sensitivity. System
sensitivity is one of the critical parameters of each scanner, depending on the detector
technology, crystal material, and axial field-of-view (FOV) in conventional cylindrical
scanners.[13 ]
[14 ] In addition, the detectability of 18 F-FDG features in PET/CT scans is influenced by reconstruction algorithms. Therefore,
the 18 F-FDG guidelines need to be updated to consider different scanner types with different
sensitivity and reconstruction algorithms. GE Healthcare has recently manufactured
GE Discovery IQ, a highly sensitive long-axial FOV PET scanner based on five-ring
BGO detectors with a sensitivity of 22 kcps/MBq.[15 ]
[16 ] In this study, we aimed to optimize the FDG-administered dose based on patient specifications
in whole-body scans using this scanner.
Methods
Patient
18 F-FDG PET/CT scans of 101 patients of both sexes (65 females and 36 males) were randomly
selected. The patients were scanned according to standard clinical protocols and guidelines
of the EANM in 18 F-FDG PET/CT imaging.[6 ] Body weight of 45 to 113 kg and different clinical indications were included. All
studies were performed retrospectively using anonymized clinical patient data. All
patients received a dose of approximately 0.1 mci (3.7 MBq) per kilogram of body weight
according to the current guidelines of the EANM, as shown in [Table 1 ]. For adherence to the guidelines for patient preparation, the scans were acquired
60 ± 5 minutes after the injection with the patients in the supine position and their
arms up. Furthermore, different time per bed positions ranging from 1.3 to 6 minutes
in terms of minutes per bed position (mpb) were used for the patients. Based on the
GE-recommended protocol, to achieve higher image quality, higher time per bed position
was used for weights greater than 60 kg.
Table 1
Patient characteristics and acquisition parameters
Number of patients
Weight (kg)
Height (cm)
Prescribed dose (MBq)
Time per bed position (min)
Range
Mean ± SD
Range
Mean ± SD
Range
Mean ± SD
Range
Mean ± SD
101 (65 female + 36 male)
45–113
72.2 ± 13
147–183
165 ± 9
114–470
305.7 ± 59
1.3–6
4.6 ± 0.7
Positron Emission Tomography/Computed Tomography Imaging
All images were scanned using the GE Discovery IQ PET/CT system (General Electric
Healthcare, WI, United States), which combines a high-sensitivity PET scanner (22
cps/kBq) and a 16-slice CT scanner (120 kV, 80 mA).[16 ] A reconstruction algorithm featuring 4 iterations, 12 subsets, and a 6.4-mm Gaussian
postprocessing filter with resolution recovery capability (OSEM + SharpIR) was used
as a routine reconstruction technique for this system. The 192 × 192 matrix size,
resulting in a 3.64 × 3.64 × 3.26-mm pixel size, formed PET images.
Quantitative Image Analysis
Parameters, such as injected activity, time per bed position, and body weight (kg),
were derived for each patient, body mass index (BMI) was calculated, and lean body
mass (LBM) was theoretically defined and calculated based on the method presented
by Hume.[17 ] The dose time product (DTP) was obtained using the formula FO, where A is the injected activity and t is the scan time (time per bed position). The SNR in the liver was selected as an
index of image quality because of the relatively homogeneous uptake of FDG. Patients
with inhomogeneous uptake mainly due to metastasis or other irregularities in the
liver were excluded from this study. A spherical voxel of interest (VOI) 40 mm in
diameter was placed in the center of the largest liver axial slice to avoid partial
volume effects at the liver edges and separately from the porta hepatis and major
vessel area of the liver to target only liver tissues using Amid software (version
1.0.3). The SNR was calculated according to Equation 1:
Mean is the mean pixel value within the VOI.
SD is the standard deviation in the observed region.
The result of the equation is reported as SNR liver (SNRL) for each patient. SNRL
was normalized to eliminate its dependency on time per bed position for each patient
(SNRnorm [MBq·min]−1/2 ) according to Equation 2 and then plotted against different patient parameters, such
as weight (kg), BMI, and lean body mass, as shown in [Fig. 1 ].
Fig. 1
Top row : Signal-to-noise ratio of the liver (SNRL) as a function of patient parameters (weight,
BMI, and LBM). Bottom row : Normalized SNR (SNR norm) for injected dose activity and scan time as a function
of patient weight, BMI, and LBM. Scattered dots are the SNR norm data fitted to nonlinear regression (dotted curve ).
Nonlinear fitting was performed on the graph of SNRnorm and patient parameters to find a and p function values in Equation 3, where SNRfit ([MBq min]−1/2 ) is the result of the fit:
p is the patient-dependent parameter.
a , d is the fitting-derived constants.
A combination of Equations 2 and 3 showed that SNRL, and hence, the image quality,
was constant if this constant is equal to the acceptable SNRL (SNRacc ).
Qualitative Image Evaluation
Two expert nuclear medicine physicians determined the SNRacc for all patient images using each algorithm. SNRacc represents the constant of SNRL corresponding to the highest value of the patient parameters for which the image
quality was still acceptable. To achieve this goal, raw image data representing a
96-kg male, 287-MBq injected dose, and 160-s scan time per bed position were selected
and then reconstructed using the following different time per bed positions (80, 40,
20,10, and 5 seconds) so that new images with different qualities was generated. All
resulting images in the coronal and axial views were evaluated by two expert nuclear
medicine physicians (more than 8 years of experience) to select the least acceptable
image qualitatively. Consequently, the image that exhibited the lowest acceptable
SNR (SNRacc) required for accurate diagnosis was chosen. It should be noted that the
physician was blinded to all patient information, such as time per bed position (the
reconstructed time) and injection activity. On the contrary, quantitatively, the noise
index of all the generated images was measured by obtaining the coefficient of variation
(COV% =FO). All images with higher SNR (higher than SNRacc) presented to have a good
coefficient of variation, by the same token they were scored a quality level of good
or moderate in the clinical visual assessment. A new DTP was calculated according
to patient-dependent parameters using Equation 4:
Finally, a new injection activity and time per bed position were calculated using
the new DTP value. It's worth to be mentioned that this method was used previously
by Groote et al to have similar outcomes, and yet his study was valid only on the
Biograph TruePoint PET-CT scanner.
Results
The measurements and calculations of patient characteristics and parameters are displayed
in [Table 1 ]. The graphs in [Fig. 1 ] show the measured SNRL as a function of body weight, BMI, and LBM (according to
the theoretical calculations). A linear function was fitted to the scatter data to
determine the behavior of SNRL against each patient's parameters.
[Fig. 1 ] demonstrates an almost linear fitting for SNR (SNRfit ), which was achieved by fitting the SNRnorm (SNRL after normalization) with all the parameters, which resulted in the determination
of the fit parameters a and d for each patient parameter. The regressions in [Fig. 1 ] were obtained using the data of all samples scanned using our scanner in this study.
The R2 values were 0.24, 0.22, and 0.18 for patient weight, BMI, and LBM respectively. Patient
weight had the highest R2 and was the easiest parameter to implement in the clinic; therefore, it was chosen
as the best parameter for image quality assessment and dose optimization.
[Fig. 2 ] shows the FDGPET images generated at different time per bed positions in coronal
and axial views. All images were shown to the physician and the 80-second duration
was selected as the time per bed position required for minimum acceptable quality
image; however, 160-second time per bed position provided much higher confidence for
the physician when reporting. For these time per bed position images, the SNR was
calculated as 7.9 and 12.3 for 80 and 160 seconds, respectively. In addition to qualitative
and quantitative assessments, the COV of each image was calculated and plotted. The
COVs for 60 and 180 seconds were 12.9 and 8.1%, respectively, and for the other scan
time per bed position,COVs were more than 15% ([Fig. 3 ]). Appropriate and acceptable COV should be under 15%;[18 ] therefore, just these two time per bed position 180 and 60 s were included, and
the others scan times were excluded.
Fig. 2 Coronal and transaxial views of a whole-body FDG-PET scan for a 95-kg patient (injected
dose = 287 MBq) with different scan times, including 5, 10, 20, 40, 80, 160 s. Scans
were performed on the Discovery IQ five-ring PET-CT.
Fig. 3 Coefficient of variation (COV) values for the liver of a 96-kg patient in whole-body
FDG-PET scan using different scan times. The threshold COV, which provides acceptable
image is 15%; therefore, COVs of 12.9% and 8.1% are included as reliable images.
[Table 2 ] illustrates the process for obtaining a new DTP. The values for the fit parameters
a and d have been shown in [Table 2 ]. A paired-sample t -test was used to show a significant difference (p < 0.0001) between the old and new DTP values. Based on the acceptable and high-confidence
SNR, the old DTP value was reduced by approximately 74 and 38%, respectively.
Table 2
The difference between the new dose time product (DTP) formula in both acceptable-
and high-confidence signal-to-noise ratio values and the old DTP formula
Index/parameter
a
d
Fitting equation
SNR acceptable
Dose time Product DTP formula
Diff in DTP%
T test DTP reduction
Weight (acceptable confidence)
5.89
0.62
5.89 X−0.62
7.9
1.8 (weight)1.24
−74.4
(p < 0.0001)
Weight (higher confidence)
5.89
0.62
5.89 X−0.62
12.3
4.3 (weight)1.24
−38.9
(p < 0.0001)
In [Fig. 4 ], the Wilcoxon matched-pairs signed-rank test showed a significant difference (p < 0.0001) between the old DTP values and the optimized new DTP values. There is a
significant reduction (p < 0.0001) in the new optimized DTP compared with the old DTP based on patient weight
([Fig. 4A ]). This reduction was more significant for an acceptable SNR. Final new DTP formula
(activity × time), depending on the patient weight was obtained as follows:
Fig. 4 (A ) The dose time product (DTP) according to the patient weight following the EANM guidelines
and routine clinical procedures (gray dots ) and the new-DTP proposed formula (black dots ) with acceptable confidence. (B ) The DTP according to the patient weight following the EANM guidelines and routine
clinical procedures (gray dots ) and the new-DTP proposed formula (black dots ) with higher confidence.
MBq for acceptable confidence.
MBq for higher confidence.
Discussion
The guidelines for tumor imaging using 18 F-FDG show an average injected activity of 370 to 740 MBq.[6 ]
[7 ] However, this dose recommendation does not consider the image reconstruction algorithms
used and also states that the dose can be lowered in highly sensitive PET/CT systems.
A highly sensitive PET/CT scan, such as a GE Discovery 5-ring BGO-based detector,
has shown a significant positive impact on the image quality. Some studies have dealt
with dose-time optimization based on patients' physical specifications. Niederkohr
et al suggested that using specific equipment, a slight reduction might be possible
in the administered FDG dose or the PET scan time per bed positions to levels below
the values identified in the EANM/SNMMI procedure guidelines.[9 ] Wickham et al reported a reduction in the mean activity administered to a group
of patients compared with the current protocol with the same consistent image quality.[10 ] Prieto et al indicated that with 18 F-FDG, an injection dose reduction of 23.4% (down to 3.57 MBq/kg) can provide an acceptable
image quality.[19 ] Nevertheless, previous studies have been performed using a four-ring lutetium oxyorthosilicate
PET/CT scanner. Dziuk et al reported that 18 F-FDG-injected dose could be reduced by up to 25% when using a five-ring BGO crystal
PET/CT camera, without substantial impact on image quality. However, this study only
considered patients with Hodgkin lymphoma.[8 ] According to the existing guidelines, advanced PET/CT technology allows for a significant
reduction in radiotracer doses. However, these studies were limited by the systems
they used and the approach they used was not adopted by other scanners. In this study,
images were acquired using a five-ring BGO-based GE discovery-IQ PET/CT scanner. This
scanner has a sensitivity of 22 cps/MBq, which is almost three times more sensitive
than that of conventional scanners. The high sensitivity of the scanner was achieved
using numerous technological modifications, including the three-dimensional mode,
an extended axial FOV, and an increase in the number of detector rings from two to
five along the FOV. The data in [Fig. 1 ] (top row ) were obtained using a linear relationship between patient parameters and FDG dose
in both algorithms, but the scan time per bed position varied for different bodyweight
classes. However, Equation 2 can be used to adjust the scan-time adaptation. Based
on other findings,[20 ] SNRL graphs, and the routine EANM guidelines, it was observed that SNRL decreases with increasing body weight and other parameters for other scanners. However,
in our scanner, we observed a slight increase for patients weighing more than 60 kg
because we increased the scan times based on GE recommendations to prevent image degradation.
For SNRL and SNR norm calculations, the method was based on a study by Groot et al
in 2014. The image quality analysis was based on the liver SNR. Furthermore, physiology
can also affect the SNR of the patients. Variations in plasma clearance, overweight,
and/or plasma glucose levels might affect the biodistribution of FDG and, thus, the
SNR. However, this study suggests that these effects are either unusual or not as
noticeable as the attenuation. The use of the liver as a reference for image quality
in clinical observations and image analysis is an acceptable method.[21 ]
[22 ]
[23 ] The liver was chosen because it is the only organ in the body that shows relatively
uniform absorption of FDG. SNRL, on the contrary, represents physiological uptake
variability. As the only exclusion criterion for this study was the heterogeneity
of liver uptake, our findings can be generalized to all FDG whole-body scans if our
scanner is used. Normalizing the SNR (the correction process of different times of
bed per position and injected activity) is a valid method for quantifying image quality
independent of time (minutes) per bed position (mbp), according to Cox et al.[24 ]
[Table 2 ] shows the process of obtaining new DTP values under these two conditions. By applying
the value of SNRacc to Equation 4, we ensured that the output (DTPnew) was within the acceptable image
quality. The new DTP corresponds to the DTP in the conventional method (FO) but considers
patient parameters and is more sensitive to the type of algorithm used for processing
images. Cox et al and de Groot et al proposed methods to obtain new DTP values within
acceptable image quality for adult patients, depending on their SNR.[20 ]
[24 ] Fitting SNRnorm to different patient-dependent parameters (p ) showed that SNRnom had the strongest relationship with body weight (the highest
R2 ). Accordingly, this would lead to a greater influence on the optimized DTP values
among the other parameters. Because weight is the simplest patient-dependent parameter
and a very practical parameter to use, the choice for body weight was considered to
be used in the optimization of the FDG-injected dose. The DTP values were tested with
a paired-samples t -test, demonstrating a dramatic decrease (p < 0.0001) in the new DTP values in the two states compared with the conventional
DTP (DTP before optimization) values. Our proposed formula for injected dose can significantly
reduce the dose received by the patient. On the contrary, based on our new DTP, if
we want to use conventional injection parameters, we can reduce the scan time, which
in turn can decrease the artifacts due to patient movements and increase the PET-CT
center throughput.
Conclusion
Compared to BMI and LBM, patient weight is the best parameter with the highest R2 and is easy to use in the clinic for 18 F-FDG PET/CT image quality assessment. The new FDG dose regimen based on the patient's
body weight is recommended for new generations of highly sensitive scanners. For our
highly sensitive BGO PET-CT scanner (Discovery IQ 5 ring), we proposed a new dose-time
regimen based on body weight that can significantly reduce the injection dose and
scan times while maintaining sufficient image quality for diagnosis.