Abstract
Pre- and posttreatment PET comparative scans should ideally be obtained with identical acquisition and processing, but this is often impractical. The degree to which differing protocols affect PERCIST classification is unclear. This study evaluates the consistency of PERCIST classification across different reconstruction algorithms and whether a proprietary software tool can harmonize SUV estimation sufficiently to provide consistent response classification. Methods: Eighty-six patients with non–small cell lung cancer, colorectal liver metastases, or metastatic melanoma who were scanned for therapy monitoring purposes were prospectively recruited in this multicenter trial. Pre- and posttreatment PET scans were acquired in protocols compliant with the Society of Nuclear Medicine and Molecular Imaging and the European Association of Nuclear Medicine (EANM) acquisition guidelines and were reconstructed with a point spread function (PSF) or PSF + time-of-flight (TOF) for optimal tumor detection and also with standardized ordered-subset expectation maximization (OSEM) known to fulfill EANM harmonizing standards. After reconstruction, a proprietary software solution was applied to the PSF ± TOF data (PSF ± TOF.EQ) to harmonize SUVs with the OSEM values. The impact of differing reconstructions on PERCIST classification was evaluated. Results: For the OSEMPET1/OSEMPET2 (OSEM reconstruction for pre- and posttherapeutic PET, respectively) scenario, which was taken as the reference standard, the change in SUL was −41% ± 25 and +56% ± 62 in the groups of tumors showing a decrease and an increase in 18F-FDG uptake, respectively. The use of PSF reconstruction affected classification of tumor response. For example, taking the PSF ± TOFPET1/OSEMPET2 scenario increased the apparent reduction in SUL in responding tumors (−48% ± 22) but reduced the apparent increase in SUL in progressing tumors (+37% ± 43), as compared with the OSEMPET1/OSEMPET2 scenario. As a result, variation in reconstruction methodology (PSF ± TOFPET1/OSEMPET2 or OSEM PET1/PSF ± TOFPET2) led to 13 of 86 (15%) and 17 of 86 (20%) PERCIST classification discordances, respectively. Agreement was better for these scenarios with application of the propriety filter, with κ values of 1 and 0.95 compared with 0.79 and 0.72, respectively. Conclusion: Reconstruction algorithm–dependent variability in PERCIST classification is a significant issue but can be overcome by harmonizing SULs using a proprietary software tool.
PET with 18F-FDG is increasingly being used for response evaluation in cancer patients, especially in clinical trials (1). For pragmatic reasons, SUV is the most frequently used quantitative parameter. In an effort to bring consistency to the classification of response across trials, emulating the use of the RECIST for radiologic examinations (2), various schema based on the degree of SUV change after treatment have been proposed. The European Organization for Research and Treatment of Cancer criteria (3) have been largely supplanted by PERCIST (4).
However, many sources of error in SUV measurement exist (5,6). In particular, technologic improvements can lead to significant device-dependent and reconstruction-dependent variations in quantitative values. For instance, point spread function (PSF) reconstruction, which improves spatial resolution throughout the entire field of view (7) and which is commercially available from all major PET vendors, has been shown to increase SUV by up to 66% compared with conventional OSEM reconstruction (8). Although this might be advantageous for the detection of small lesions and increase clinical confidence when interpreting scans, this could lead to classification errors by exceeding thresholds used for discriminating between responding and nonresponding tumors unless acquisition and processing of pre- and posttreatment scans are performed on the same scanner and processed identically. However, in busy nuclear medicine departments, which may have several scanners or which periodically update equipment, and with patient mobility requiring scanning at different sites, this may not always be practical.
Therefore, there is a growing interest in SUV harmonization strategies such as the European Association Research Ltd. accreditation program (9), the North American Quantitative Imaging Biomarkers Alliance (10), and the Uniform Protocols for Imaging in Clinical Trials (11), which aim at minimizing the variability in SUV measurements by harmonizing patient preparation and scan acquisition and processing (12). Although many sources of error in SUV measurements can be overcome by complying with EANM or Society of Nuclear Medicine and Molecular Imaging guidelines for PET tumor imaging (13–15), reconstruction-dependent variations require either the use of an additional filtering step (16) or the generation of 2 sets of images, 1 set to provide optimal diagnostic quality and 1 to meet quantitative harmonization standards (17). However, the impact of reconstruction methodology on classification of treatment response is unclear.
To assess this impact, we reconstructed the same PET raw data with an OSEM algorithm known to meet EANM requirements and also with PSF with or without TOF reconstruction (PSF ± TOF). Postreconstruction filtering was then applied to the PSF or PSF plus TOF reconstruction with EQ.PET, a proprietary software solution allowing visualization of optimized images while simultaneously obtaining harmonized SUVs (16).
The primary aim of this multicenter prospective study was to evaluate the impact of SUV reconstruction-dependency on PERCIST classification and the ability of EQ.PET technology to minimize variability in response assessment.
MATERIALS AND METHODS
Patients
Over an 18-mo period, patients with non–small cell lung cancer, colorectal liver metastases, or metastatic melanoma and scanned for monitoring efficacy of chemotherapy, molecularly targeted therapies, or radiotherapy were prospectively included in 3 PET centers. Informed consent was waived for this type of study by the local ethics committee (Ref A12-D24-VOL13, Comité de protection des personnes Nord-Ouest III) because the scans were obtained for clinical indications and the trial procedures were performed independently without influencing clinical reporting.
PET Systems
Data from the following 3 PET systems were used for this study: a Biograph 6 TrueV with PSF reconstruction, an mCT with PSF + TOF, and a Biograph 64 TrueV with PSF reconstruction (Siemens Medical Solutions). Both the Biograph systems were equipped with an extended axial field of view.
Patient Preparation
All patients were requested to fast for 6 h before the 18F-FDG injection. Patient height, weight, and blood glucose levels were recorded. Patients were injected intravenously with 18F-FDG, followed by a 60-min rest in a warm room. The injected activity and the exact delay between injection and the start of the acquisition were extracted from the DICOM standard headers and were recorded for each patient for baseline scan (PET1) and posttherapeutic scan (PET2).
PET Acquisition and Reconstruction Parameters
A daily calibration of each PET system was performed with a 68Ge source according to the manufacturer’s protocol. A quarterly cross-calibration of each PET system was performed according to the EANM guidelines, as described elsewhere (13,14), and all clocks were synchronized weekly.
The PET acquisition was performed in 3-dimensional (3D) mode. Scatter and attenuation corrections were applied on all PET acquisitions. Patients were scanned from skull vertex or base to the mid thighs, with acquisition extended to legs in melanoma patients with primary site of disease in the lower limbs.
All raw PET data were reconstructed with the local PSF ± TOF settings for optimal lesion detection and a 3D OSEM reconstruction algorithm fulfilling the EANM guidelines regarding recovery coefficients (Table 1).
PET/CT Acquisition and Reconstruction Parameters for the 3 Participating Centers
EQ.PET Technology
For each PET system, the EQ.PET filter was calculated on the phantom data of each PSF ± TOF reconstruction as described in detail elsewhere (17). Briefly, the recovery coefficients (defined as the ratio between the measured and true activity concentration for each sphere) of a National Electrical Manufacturers Association NU2 phantom scanned per EANM guidelines were aligned to the EANM reference recovery coefficients by applying a gaussian filter.
PERCIST Evaluation
All PET examinations were analyzed with Syngo.via software equipped with EQ.PET (Siemens Medical Solutions). For interpretation purposes, both the reconstruction for optimal lesion detection (PSF ± TOF) and the OSEM reconstruction were displayed on the screen together with the EQ.PET-filtered harmonized SUV results for the tumor regions of interest and the liver background. The EQ.PET-filtered images were not displayed on the screen.
As defined in PERCIST (4), the measurable target lesion is the single most intense tumor site on pre- and posttreatment scans, which means that the target lesion is not necessarily the same before and after treatment.
To evaluate the impact of reconstruction methods on assessment of therapeutic response, we first assessed the accuracy of EQ.PET in correcting PSF-reconstructed lesions by assigning the same lesion (i.e., the most intense lesion determined on OSEM pre- and posttreatment scans) as a target for all reconstructions (Fig. 1A). These results are presented in Figure 2. Subsequently, we evaluated the impact of reconstruction inconsistency on PERCIST classification in real life by assessing whether the use of a different reconstruction algorithm for pre- and posttreatment scans would lead to a change in the most intense lesion location with consequent impact on PERCIST classification. For this analysis, we selected the most intense lesion on each reconstruction (Fig. 1B). Finally, we studied whether using PSF ± TOF reconstruction for both the pre- and the posttreatment scans would give results similar to those using the former-generation OSEM algorithm (Fig. 1C).
Flow chart describing different scenarios for determining accuracy of EQ.PET technology in setting of therapy assessment with PERCIST.
(A and B) Relationship between SULmean in tumor lesions extracted from PSF ± TOF or PSF ± TOF.EQ and OSEM images, assessed using Bland–Altman plots. (C) Mean ± SD ratio of SULpeak obtained with conventional OSEM algorithm and those obtained with PSF or PSF + TOF reconstructions are shown before and after application of EQ.PET technology. (D and E) Ratio of PSF ± TOF.EQ and OSEM SULpeak depending on target lesion size and tumor/background.
In practice, the most intense lesion was located by scaling the 3D maximum-intensity-projection view on both the OSEM and the PSF ± TOF reconstructions. The location of this lesion for both reconstructions was noted. Then, volumes of interest were drawn on one reconstruction and automatically propagated to the second set of reconstructions (propagation from OSEM to PSF ± TOF and vice versa). Within these volumes of interest, lean body mass SUVpeak (SULpeak) was measured. Background activity (SULmean) was measured in an automatically placed 3-cm-diameter sphere in the right liver lobe avoiding metastases, especially in colorectal cancer patients, and in a 1-cm-diameter and 2-cm-height cylinder in the descending thoracic aorta.
Additionally, a volume of interest (1-cm-diameter and 2-cm-height cylinder) was placed next to the target lesion to assess tumor-to-background ratios.
On the basis of SULpeak variation between the pre- and posttreatment scan, patients were classified according to PERCIST as follows: complete metabolic response (CMR)—complete resolution of 18F-FDG uptake in the tumor volume, with tumor SUL lower than liver SUL and background blood pool, and disappearance of all lesions if multiple; partial metabolic response (PMR)—at least 30% reduction in tumor uptake; stable metabolic disease (SMD)—less than 30% increase or less than 30% decrease in tumor 18F-FDG SULpeak and no new lesions; and progressive metabolic disease (PMD)—greater than 30% increase in 18F-FDG tumor SULpeak within the tumor or appearance of new lesions.
Statistical Analysis
Quantitative data from clinical PET/CT examinations are presented as mean ± SD. The relationship between PSF ± TOF, PSF ± TOF.EQ, and OSEM quantitative values was assessed with Bland–Altman plots. The PSF ± TOF/OSEM SULpeak ratios and PSF ± TOF.EQ/OSEM SULpeak ratios were compared using the Wilcoxon test for paired samples. To assess potential confounding factors, the ratios between PSF ± TOF.EQ and OSEM SULpeak according target lesion size (<10, 10–20, and >20 mm), tumor-to-background ratio (<5, 5–10, and >10), and reconstruction algorithm (PSF or PSF + TOF) were compared using the Kruskal–Wallis test (with a post hoc Dunn test) for multiple-groups comparison or the Mann–Whitney test for unpaired samples when appropriate. A 2-tailed P value of less than 0.05 was considered statistically significant.
Levels of agreement between the different types of reconstruction were evaluated using the κ-statistic. OSEM reconstruction both for pre- and for posttherapeutic PET examination (OSEMPET1/OSEMPET2) was used as the current standard to classify the therapeutic response of each lesion and compared with other scenarios described in Figure 1. κ-values were reported using the benchmarks of Landis and Koch (18).
Graphs and analyses were performed using Prism GraphPad and the Vassar University website for statistical computation (http://vassarstats.net).
RESULTS
Patients’ Demographics
The patients’ sex ratio (male-to-female) was 2.4, and mean age (±SD) was 63 ± 9 y. Details about the different treatment modalities for the 3 different cancer types can be found in Table 2. The interval between the pre- and posttreatment PET scans was 114 ± 55 d.
Treatment Modalities
Compliance to PERCIST Acquisition Requirements
The injected activity (MBq/kg) was 3.96 ± 0.29 and 3.94 ± 0.24 for pre- and posttreatment scans, respectively. The percentage variation between injected doses for the pre- and posttreatment scans ranged from 0% to 38% and exceeded the 20% limit in only 3 cases. PERCIST requirements were met in 83 of 86 patients (97%).
The uptake time was 64 ± 5 and 64 ± 7 min for the pre- and posttreatment scans, respectively. The absolute difference in uptake time between the pre- and posttreatment scans ranged from 0 to 42 min and exceeded the 20-min limit in only 3 cases. PERCIST requirements were met in 83 of 86 patients (97%).
In each individual patient, the baseline and the follow-up scans were always acquired on the same PET/CT system. When the 2 centers in which acquisition parameters in terms of time per bed position and postfiltering differed depending on patient weight or body mass index were considered, interscan acquisition consistency was achieved in 81 of 86 (94%) patients.
Ability of EQ.PET Technology to Harmonize SUL Assessments
The mean SULpeak (±SD) for OSEM, PSF ± TOF, and PSF ± TOF.EQ reconstructions were 7.5 ± 5.6, 8.4 ± 6.2, and 7.6 ± 5.7, respectively.
The mean ratio between PSF ± TOF and OSEM reconstructions for SULpeak was 1.13 (95% confidence interval, 0.90–1.36) (Fig. 2A). After application of the EQ.PET filter, this ratio was reduced to 1.03 (95% confidence interval, 0.94–1.12) (Fig. 2B).
The impact of potential confounders on the ratios between PSF ± TOF.EQ and OSEM reconstructions for SULpeak are shown in Figures 2C–2E. Because no difference in the ratios of SULpeak was observed between the PSF and PSF + TOF reconstructions, we decided not to separate these results in subsequent analyses. The ratios of SULpeak were not influenced by the size of the target lesion or by the background surrounding the target lesion.
Impact of Reconstruction-Dependent Variation of SUL on PERCIST Evaluation
The target lesion selected was the same for OSEM and PSF ± TOF reconstructions for 83 (97%) of the pretreatment scans and 85 (99%) of the posttreatment scans. The mean size of the target lesion was 31 ± 21 mm for the baseline scan and 27 ± 22 mm for the posttreatment scan.
The variation in SULpeak between the pre- and posttreatment scans is shown in Figure 3. For the OSEM/OSEM scenario, which was taken as the reference standard, the change in SUL was −41% ± 25 and +56% ± 62 in the groups of tumors showing a decrease and an increase in 18F-FDG uptake, respectively. The use of PSF reconstruction affected classification of tumor response, depending on whether this reconstruction was used for the pre- or posttreatment scans. For example, the PSF ± TOFPET1/OSEMPET2 scenario increased the apparent reduction in SUL in responding tumors (−48% ± 22) and reduced the apparent increase in SUL in progressing tumors (+37% ± 43), as compared with the OSEMPET1/OSEMPET2 scenario described above. Accordingly, inconsistent reconstructions induced discordant response classifications among the different scenarios.
Impact of reconstruction consistency on percentage variation in SULpeak in responding (left) and progressing (right) tumors. Data are mean ± SD.
When OSEM for the pre- and posttreatment scans was used, PET classified 10 patients as CMR, 21 as PMR, 26 as SMD, and 29 as PMD (Fig. 4; Supplemental Table 1 [supplemental materials are available at http://jnm.snmjournals.org]). CMR occurred in 6 patients with a decrease in SULpeak to a level below the liver and blood-pool background and in 4 patients to complete disappearance of the target lesions. PMD occurred in 7 patients with an increase in tumor SULpeak greater than 30% and in 22 patients with new lesions on the posttreatment scan.
Impact of reconstruction consistency on PERCIST classification. PERCIST classification is shown for standard of reference and for other scenarios: reconstruction inconsistency between baseline and posttreatment scans (A), use of EQ.PET methodology either for baseline or for posttreatment scan (B), and reconstruction consistency but use of different generations of reconstruction algorithms (C).
Agreement levels between the OSEMPET1/OSEMPET2 scenario and other scenarios involving reconstruction inconsistency (Table 3) were found to be almost perfect, with narrow confidence intervals for the scenarios using EQ.PET-filtered data either before or after treatment and the reconstruction-consistent scenario PSF ± TOFPET1/PSF ± TOFPET2. Agreement levels were fair to substantial for the scenario OSEMPET1/PSF ± TOFPET2 and PSF ± TOFPET1/OSEMPET2, with wide confidence intervals.
Agreement Levels Between OSEMPET1/OSEMPET2 Scenario and Other Scenarios Involving Reconstruction Inconsistency
Table 4 shows the number of discordances in the PERCIST classification that occurred for the different scenarios tested. Discordances (n = 42) were most frequent for the scenarios OSEMPET1/PSF ± TOFPET2 (n = 17) and PSF ± TOFPET1/OSEMPET2 (n = 13). For OSEMPET1/PSF ± TOFPET2, these led to 7 patients being classified as SMD instead of PMR, 3 as PMR instead of CMR, and 7 as PMD instead of SMD. For PSF ± TOFPET1/OSEMPET2, 11 patients were classified as PMR instead of SMD and 2 as SMD instead of PMD. Figure 5 illustrates a patient classified as SMD according to the OSEMPET1/OSEMPET2 standard of reference, whereas reconstruction inconsistency (OSEMPET1/PSF + TOFPET2) led to a PMD classification.
Number of Discordances in PERCIST Classification That Occurred for Different Scenarios Tested
A 50-y-old man with liver metastasis (right lobe) from colon cancer treated by chemotherapy, classified as SMD according to OSEMPET1/OSEMPET2 standard of reference whereas reconstruction inconsistency (OSEMPET1/PSF + TOFPET2, a scenario mimicking system upgrade during a trial) led to PMD classification. Use of EQ.PET technology correctly classified patient as SMD. (A) Transverse slices at level of the liver metastasis for OSEM and PSF + TOF reconstructions and for baseline and posttreatment scans. (B) Percentage change in SULpeak and PERCIST classification according to different scenarios.
Impact of Reconstruction-Dependent Variation of SUL on Liver Reference Background
As shown in Figure 6, the PERCIST requirement that normal liver SUL must be within 20% and <0.3 SUL mean units for the pre- and posttreatment PET scans to be assessable was not met for 15 (17%) for OSEMPET1/OSEMPET2 and up to 17 (20%) of the 86 patients for the other scenarios of reconstruction inconsistency.
Absolute (A) and relative (B) variation in liver SULmean within pre- and posttreatment scans. Red lines illustrate PERCIST requirements regarding these parameters.
In none of the patients was this due to a difference in uptake time > 20 min, or variation in injected dose > 20% between the pre- and posttreatment PET scans. In 3 patients, there was a change in blood glucose level of > 1 mmol/, and 2 patients experienced a weight loss > 5 kg.
DISCUSSION
18F-FDG PET is being increasingly used for therapy monitoring in cancer patients. In the context of constant technologic evolutions that affect PET quantitation, harmonization strategies have emerged (9,10,12). In a previous study, we investigated the capability of the EQ.PET technology to harmonize PSF ± TOF and OSEM SUV metrics while optimizing tumor detection (16). In this prospective multicenter study, we evaluated the impact of inconsistent reconstruction on PERCIST response classification, demonstrating variation in up to 20% of cases. Further, we showed that proprietary software, such as the EQ.PET technology, provided more consistent response classification. The EQ.PET technology is not affected by the type of reconstruction, the tumor size, or the tumor-to-background level. This has practical advantages when use of the same scanner for both scans is impractical, or when there is inadvertent variation of acquisition/reconstruction settings. The latter situation seems relatively common even in centers running the same PET system, as recently reported by the Clinical Trials Network of the Society of Nuclear Medicine and Molecular Imaging (19).
In centers running 2 or more PET systems, this problem is compounded as illustrated by the study by Skougaard et al. (20), in which 12 of 81 (14%) patients undergoing pre- and posttreatment PET in the same department were excluded for analysis because they were scanned on 2 different generation PET systems.
We observed compliance in more than 90% of cases to PERCIST requirements regarding the injected dose and the uptake time between the pre- and posttreatment scans in a series of patients scanned in routine clinical practice. Despite good acquisition compliance, inconsistent reconstruction led to variability in PERCIST classification compared with OSEM as the reference standard. Taking, for example, the scenario of a system upgrade during a trial, the use of OSEM for the pretreatment scan while using PSF ± TOF for the posttreatment scan led to discordant response assessments in 17 of 86 (20%) (Table 4). It is noteworthy that a change in selected PERCIST target lesion occurred in only 3 of 172 scans (2%) and that among patients classified as PMD because of the appearance of new lesions, OSEM and PSF ± TOF performed equally in detecting these new lesions despite the potential for PSF reconstruction to detect smaller cancer lesions and to significantly increase SUV metrics as compared with OSEM reconstruction (8).
We found that with the appropriate EQ.PET filter for each center’s PSF ± TOF reconstruction, we were able to harmonize PET quantitative data for tumors with a mean ratio of 1.02 for SULpeak, with narrow confidence interval. The use of the EQ.PET methodology for either the pre- or the posttreatment scans gave almost perfect agreement levels in comparison with the OSEMPET1/OSEMPET2 reference standard, with narrow confidence intervals. We observed only 3 discordances for the OSEMPET1/PSF ± TOF.EQ PET2 versus OSEMPET1/OSEMPET2 scenario, and no discordance occurred for the PSF ± TOF.EQ PET1/OSEMPET2 versus OSEMPET1/OSEMPET2 scenario. In the melanoma group, the OSEMPET1/OSEMPET2 versus OSEMPET1/PSF ± TOF PET2 scenario led to a 0.58 κ-value that can be explained by 3 discordances (patients classified as SMD for the OSEMPET1/OSEMPET2 scenario and classified as PMD for the OSEMPET1/PSF ± TOFPET2 scenario). This involved a group of only 11 patients, contributing to the wide confidence interval.
Consistent reconstruction (i.e., the PSF ± TOFPET1/PSF ± TOFPET2 and PSF ± TOF.EQPET1/PSF ± TOF.EQPET2) did not, however, give perfect agreement compared with the OSEMPET1/OSEMPET2 standard of reference. These discordances were due to differences in background level (liver and blood pool) on posttreatment scan among the different reconstructions, leading to CMR being changed to PMR and vice versa (Supplemental Fig. 1), and to a percentage change in SULpeak close to +30% or −30% for the OSEMPET1/OSEMPET2 scenario, resulting in changes from SMD to either PMR or PMD and vice versa for other scenarios.
When strictly applying the PERCIST criterion about normalization of liver uptake (normal liver SUL must be within 20% and <0.3 SUL mean units for PET1 and PET2 to be assessable), despite excellent compliance with acquisition consistency, a somewhat unexpected finding of this study was that 17% of the response evaluations in this study would not have been considered assessable. As this was observed for the OSEMPET1/OSEMPET2 scenario and consistent in all but 2 patients for the other scenarios, we don’t think that this is due to inconsistent reconstruction and may warrant reconsideration of this criterion, particularly for therapies that may alter hepatic metabolism. Factors influencing hepatic 18F-FDG uptake have been found to be chemotherapy, patient weight/BMI, blood glucose level, and hepatic steatosis (21). In this study, most of the 86 patients were treated with chemotherapy. However, only a few of the patients suffered from severe weight loss or unstable glycaemia.
A limitation of this study is that EQ.PET is a software solution developed for, and applied only to, scanners and reconstruction algorithms of the company that developed this product (including 2 PET/CT systems using a similar PET component) and has not been validated for equipment from other manufacturers.
The alternative approach of obtaining a second reconstruction dataset, as recommended by the European Association Research Ltd. accreditation program for quantitation, can be easily implemented in any PET unit irrespective of their equipment. Using the EQ technology to process images acquired on non-Siemens PET systems would require the vendor-neutral capacity of this software to be validated, using clinical data and other tools such as the digital reference object technique recently published by Pierce et al. (22). However, the EQ.PET software does not require a second standardized reconstruction and could be applied to oldest examinations, acquired and stored before the era of PET standardization programs, provided other sources of SUL variability are controlled and data regarding calibration of the PET system are available. Also, the EQ.PET filter could be adapted to meet any given standard, which is important in the context of evolving guidelines.
CONCLUSION
Reconstruction algorithm–dependent variability in PERCIST classification is a significant issue but can be overcome by harmonizing SULs using a proprietary software tool. Other manufacturers are encouraged either to emulate this solution or to produce a vendor-neutral approach.
DISCLOSURE
The costs of publication of this article were defrayed in part by the payment of page charges. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734. Prof. Nicolas Aide received a research grant from Siemens Molecular Imaging. No other potential conflict of interest relevant to this article was reported.
Footnotes
↵* Contributed equally to this work.
Published online Jun. 9, 2016.
- © 2016 by the Society of Nuclear Medicine and Molecular Imaging, Inc.
REFERENCES
- Received for publication January 14, 2016.
- Accepted for publication May 6, 2016.