A personal acquisition time regimen of 68Ga-DOTATATE total-body PET/CT in patients with neuroendocrine tumor (NET): a feasibility study

Background The injection activity of tracer, acquisition time, patient-specific photon attenuation, and large body mass, can influence on image quality. Fixed acquisition time and body mass related injection activity in clinical practice results in a large difference in image quality. Thus, this study proposes a patient-specific acquisition time regimen of 68 Ga-DOTATATE total-body positron emission tomography-computed tomography (PET/CT) to counteract the influence of body mass (BM, kg) on image quality, and acquire an acceptable and constant image of patients with neuroendocrine tumors (NETs). Methods The development cohort consisting of 19 consecutive patients with full activity (88.7–204.9 MBq, 2.0 ± 0.1 MBq/kg) was to establish the acquisition time regimen. The liver SNR (signal-to-noise ratio, SNRL) was normalized (SNRnorm) by the product of injected activity (MBq) and acquisition time (min). Fitting of SNRnorm against body mass (BM, kg) in linear correlation was performed. Subjective assessment of image quality was performed using a 5-point Likert scale to determine the acceptable threshold of SNRL, and an optimized acquisition regimen based on BM was proposed, and validated its feasibility through the validation cohort of 57 consecutive NET patients with half activity (66.9 ± 11.3 MBq, 1.0 ± 0.1 MBq/kg) and a fixed acquisition time regimen. Results The linear correlation (R2 = 0.63) between SNRnorm and BM (kg) was SNRnorm = -0.01*BM + 1.50. The threshold SNRL of acceptable image quality was 11.2. The patient-specific variable acquisition time regimen was determined as: t (min) = 125.4/(injective activity)*(-0.01*BM + 1.50)2. Based on that proposed regimen, the average acquisition time for acceptable image quality in the validation cohort was 2.99 ± 0.91 min, ranging from 2.18 to 6.35 min, which was reduced by 36.50% ~ 78.20% compared with the fixed acquisition time of 10 min. Subjective evaluation showed that acceptable image quality could be obtained at 3.00 min in the validation group, with an average subjective score of 3.44 ± 0.53 (kappa = 0.97, 95% CI: 0.96 ~ 0.98). Bland–Altman analysis revealed good agreement between the proposed regimen and the fixed acquisition time cohort. Conclusion A patient-specific acquisition time regimen was proposed in NET patients in development cohort and validated its feasibility in patients with NETs in validation cohort by 68 Ga-DOTATATE total-body PET/CT imaging. Based on the proposed regimen, the homogenous image quality with optimal acquisition time was available independent of body mass. Supplementary Information The online version contains supplementary material available at 10.1186/s40644-022-00517-8.


Conclusion:
A patient-specific acquisition time regimen was proposed in NET patients in development cohort and validated its feasibility in patients with NETs in validation cohort by 68 Ga-DOTATATE total-body PET/CT imaging. Based on the proposed regimen, the homogenous image quality with optimal acquisition time was available independent of body mass.
Total-body PET detector crystals with a size of 2.76 × 2.76 × 18.0 mm 3 coupled to silicon photomultipliers (SIPMs) shows ultrahigh sensitivity and spatial resolution [5]. Recently, a series of studies reported that low injection activity of 18 F-FDG or a shorter scan time of total-body PET/CT imaging could still be feasible for good image quality [6][7][8][9]. Additionally, influence factors of image quality were varied and complicated, such as PET equipment, activity of the tracer, acquisition time, and patient-specific photon attenuation, particularly for large body mass. In terms of conventional PET equipment, increasing injection activity and prolonging acquisition time to some extent might improve image quality. However, the probability of radiation-related injury and later-occurring effects, and the possibility of motion artifacts inevitably increased [10]. As such, fixed acquisition time in clinical practice may result in significant differences in image quality between patients with varied body mass.
Regarding constant and acceptable image quality, we previously investigated the influence of patient size on image quality, and proposed a dose regimen based on body mass index (BMI, kg/m 2 ), demonstrating the feasibility of constant image quality for 18 F-FDG total-body PET/CT [11]. In addition, adjusting the duration time per bed based on scanner sensitivity and patient-specific attenuation might acquire uniform image noise or homogenous image [12]. For 68 Ga-DOTATATE imaging, body mass (BM) was regarded as the strongest correlation with image quality [13]. However, the injection activity and acquisition time of 68 Ga-DOTATATE vary in the literature, with reduced comparability of image quality between different studies. The regimen of a variableacquisition time of 68 Ga-DOTATATE PET/CT for an acceptable and constant image has not been investigated thus far. Therefore, the aim of the present study was to propose a variable acquisition time regimen to balance the influence of scanners and BM and to obtain homogeneous image quality for patients with NETs.

Patient population
All patient information was obtained in accordance with institutional ethical standards, and all included patients waived written informed consent prior to recruitment into the study (Approval No. B2020-186R). In total, 101 consecutive patients diagnosed with or suspected of having NET who underwent 68 Ga-DOTATATE total-body PET/CT from September 2020 to January 2021 in our center were retrospectively analyzed. Prior to injection, patients were randomly and blindly (patient and imageevaluation physician) divided into the two cohorts: the development cohorts and the validation cohort. The development cohort was the full-activity group with activity of 2.0 ± 0.1 MBq/kg (88.7-204.9 MBq), and the development cohort was the low-activity group with activity of 1.0 ± 0.1 MBq/kg (66.9 ± 11.3 MBq). Figure 1 shows the process of patient recruitment, including the inclusion and exclusion criteria. We exclude patients with continuous treatment with Octreotide (N = 6), treatment with high intensity focused ultrasound in liver within 1 week (N = 1), diffused lesions in liver, fatty liver, and cirrhosis (N = 5). Finally, 48 underwent surgical resection as initial treatment, and the other 28 underwent biopsy. All cases were graded according to AJCC 2017 [14]; the remaining portion of the lesion without pathological examination was proven to be NET according to followup clinical and imaging data. All participants included in this study followed the standard procedures established by our center. 68 Ga-DOTATATE PET/CT imaging and image reconstruction 68 Ga-DOTATATE was synthesized in-house according to a method previously described in the literature [15]. None of the participants needed to fast before tracer injection. Imaging was acquired using the following steps, as illustrated in Supplement Fig. 1: (1) Low radiation dose CT was performed before PET imaging for attenuation correction with a voltage of 120 kV and a current of 10 mA. (2) Images were acquired for 10 min in 3D-list mode for 50 min post injection of 68 Ga-DOTATATE. All imaging was performed on uEXPLORER (United Imaging Healthcare, Shanghai, China) PET/CT with an axial FOV of 194 cm. (3) The tube voltage of diagnostic CT was 120 kV, and the tube current modulation technology was utilized to minimize the radiation dose.
Injection activity followed the BM-based linear dose regimen in both cohorts. Table 1 summarizes the current scanning and reconstruction protocol. In the development cohort, the images were further split into

Qualitative image analysis
The image quality of all 76 patients was independently evaluated by two nuclear medicine physicians (Jie Xiao with 3 years and Xiuli Sui with 2 years of experience in interpreting PET images); to minimize bias, they were blinded to the patient's medical history, injection activity and the reconstruction time. Two reading sessions separated by 3-4 weeks were performed by each reader. Before interpreting the images, the two readers performed consistency training using standard images formulated according to the rule of 5-points Likert scale; the intraand inter-reader agreement of these standard images should have kappa values over 0.85 (Table S1). The image quality was scored from 3 perspectives: the overall impression of the image quality, the image noise, and the lesion detectability. Score were based on a 5-point Likert scale, as follows: Score 1, image with non-diagnostic quality, excessive noise, or unfavorable lesion contrast; Score 2, acceptable image but with sub-optimal noise and lesion depiction leading to impaired diagnostic confidence; Score 3, image with quality equivalent to those used in clinical practice; Score 4, image with quality superior to the average image quality; Score 5, image with excellent quality, optimal noise, sharp lesion depiction, and free of artifact, providing diagnosis with full confidence [11]. A score of 3 was deemed as acceptable image quality in routine clinical practice in our center. IN the event of large evaluation differences between the readers, the images were discussed in a consensus meeting.

Semiquantitative image quality
To measure background uptake of the liver, an ROI (region of interest) with a diameter of 20 mm was drawn in the right lobe of the liver at the portal vein bifurcation section, avoiding any lesions and large vessels for measuring background uptake of the liver. The ROI of the lesion was placed in the highest pixel value on the transverse view, and an overall maximum of five lesions per patient was measured in multiple lesions. For objective evaluation of image quality, the signal-to-noise ratio of the liver (SNR L ) was used and calculated by dividing the liver SUV mean by its SD (Eq. 1) [16].
The dose-time product (DTP, MBq·min) could define as the product of injected activity (MBq) and acquisition time (min). The SNR norm is normalized to SNR L , which is calculated as SNR L divided by the square root of the DTP and can be assumed to be independent of the injected activity and acquisition time (Eq. 2) [13]. Therefore, SNR norm (1/sqrt (MBq·min)) can be regarded as a function of BMdependent parameters.
Linear fitting was performed with the SNR norm vs. patient BM. The mean acceptable SNR L (SNR acc ) was obtained by calculating the mean value of SNR L from all the images scored with 3 points. Finally, the variable acquisition regimen was determined as follows (Eq. 3): Injected activity where injected activity has a linear correlation with BM and SNR fit is the determining function of the fit to SNR norm vs. BM.
The variable-acquisition regimen was validated in a new cohort of 57 patients. SNR L was calculated using Eq. 1. Noise can influence the detectability of lesions, which is described as the coefficient of variation (CV): The tumor-liver ratio (TLR) and tumor-mediastinal blood pool-ratio (TMR) were calculated by dividing the SUV max of the lesion by the SUV mean of the liver and the SUV max of the lesion by the SUV mean of the ascending aorta: After evaluation of objective and subjective image quality in the validation group, the acquisition time for acceptable image quality was determined. The consistency of acquisition time between the proposed acquisition time regimen and that of the validation group was analyzed.
SUVmax of lesion SUVmean of ascending aorta

Statistical analysis
All statistical analyses were performed using IBM SPSS Statistics Version 26 (IBM Inc., Chicago, IL, USA) and Prism 8 (GraphPad Software Inc., San Diego, California, USA). Data are described as the mean ± SD. Differences in quantitative variables were assessed by analysis of variance (ANOVA) with post hoc Bonferroni adjustment for pairwise comparisons. Categorical variables were compared using the chi-square test. Cohen's kappa analysis of overall image quality was performed to evaluate inter-reader and intrareader agreement. Bland-Altman analysis was applied to determine agreement between the proposed variable acquisition time regimen and validation group data. Results were considered statistically significant if the p value was less than 0.05.

Patient characteristics
The characteristics of the two cohorts are displayed in Table 2. A total of 76 NET patients were enrolled. The development cohort consisted of 19 patients (8 male, 11 females, mean age 51.7 ± 13.7 years old, ranging from 30.0 to 74.0 y) scheduled for full activity (1.9 ± 0.2 MBq/ kg, ranging from 1.6 to 2.3 MBq/kg). The validation cohort consisted of 57 patients (36 males and 21 females; mean age, 53.6 ± 11.2 years, ranging from 30.0 to 80.0 y). The injection dose regimen was 1.0 ± 0.1 MBq/ kg, ranging from 0.8 to 1.2 MBq/kg. In addition, the mean injected activity was 120.4 ± 27.2 MBq (range:

Development of the variable acquisition time regimen
There was no significant difference in SNR L for D2-D10, though a significant difference in D15s, D30s, D45s, and D1 compared with D10 was observed. The SNR L increased with acquisition time, with a significant difference from that at D10 (Fig. 2a and Table 4). The SNR norm , i.e., the normalized SNR L , was not significantly different among the reconstruction groups (Fig. 2b). The SNR norm was then fitted with BM using a linear method with a coefficient of determination of 0.63, as illustrated in Fig. 3 (R 2 = 0.63). Therefore, the linear fit function was.
The results of subjective evaluation of image quality are presented in Table 3. The average interreader and intrareader overall image quality showed excellent agreement (all kappa > 0.85). An SNR acc value of 11.2 was obtained by calculating the average SNR L of all PET series with a score of 3 points. Thus, the variable acquisition time regimen can be deduced was Injected activity , as follows: (7) SNR norm = −0.01 * BM + 1.50 Fig. 2 The SNR L (a) and SNR norm (b) against acquisition time of full-activity cohort. ns, no significant difference    where a is the injected activity per weight (MBq/kg). Equation 8 is suitable for body mass less than 150 kg. The semiquantitative parameters are summarized in detail in Table 4. There were no differences in the SUV mean of the liver or the SUV mean of the mediastinal blood pool among the reconstruction groups (all p > 0.05). Compared with D10, the liver SD and mediastinal blood pool were only decreased in D15s and D30s (p < 0.05). Furthermore, the liver SNR, mediastinal blood pool SNR, liver CV, and mediastinal blood pool CV progressively differed for D15s, D30s, D45s, and D1 (p < 0.05); there was no difference from D2 to D10 (all p > 0.05).

Analysis of image quality in the validation cohort
Objective and subjective image quality were evaluated in the validation cohort (the low activity: 1.0 ± 0.1 MBq/kg). Table 5 provides an overview of the subjective scores for all reconstruction groups. The average overall image quality scores in the R1 and R2 groups were less than 3 and were considered to be nondiagnostic images. All images with a 3 min or longer  Fig. 4a, the SNR L was significantly lower in the R1, R2, and R3 groups than in the R10 group, though R4, R5, R8, and R10 showed no significant differences (p < 0.05). The CV is presented against acquisition time in Fig. 4b, and it decreased significantly with reconstruction time in R1, R2, R3 compared with R10. However, there was no obvious difference for R4, R5, R8, and R10. There were also no significant discrepancies in the lesion SUV max , SUV mean , SD, TMR, or TBR among the series of reconstructions (all p > 0.05, Figs. 4c-d and 5). Compared with D2 of the development cohort, R4 obtained equal values of SUV max and SUV mean for the liver, blood pool, and lesions (shown in Table 6). The CV of the liver and blood pool also showed equal. The TLR and TMR of R2 were comparable to those of R10. Referenced as PET images of R10, 90 SSTR-positive lesions were identified. Lesions were detected in the liver (38, 42.2%), bone (18, 20.0%), lymph nodes (15,16.7%), gastrointestinal tract (10, 11.1%), pancreas (5, 5.6%), breast (2, 2.2%), and mediastinum (2, 2.2%), and all of these lesions were clearly found in the R1-R10 groups (100%). Figure 6A shows the optimal acquisition time in the validation cohort, with an average time of 2.99 ± 0.91 min, ranging from 2.18 to 6.35 min, calculated by the regime of Eq. 8. The agreement between proposed and subjective optimal time was analyzed by Bland-Altman plots (Fig. 6b). It showed good agreement between the proposed regimen and the validation patient cohort. The mean bias between the proposed regimen and the validation patient cohort was -0.16 min, with 95% acceptable limits of -0.79 min and 0.48 min. A typical case is presented in Fig. 7.

Discussion
The 2017 European Association of Nuclear Medicine (EANM) procedure guidelines recommend that the administered activity of 68 Ga-DOTA-conjugated peptide ranges from 100 to 200 MBq, varying based on the PET system and patient size [17]. Although that recommendation provides a reference for clinical practice, such a broad injection range creates certain challenges with respect to the comparability and repeatability of images. In previous research, we established a convenient  patient-specific injection regimen of 18 F-FDG for repeatable and constant imaging. Thus, considering the influence of the total-body PET system and body mass on image quality, the present study proposes a personal variable acquisition time regimen to gain constant image quality and avoid extending acquisition time based on objective and subjective image quality evaluation in the development of 68 Ga-DOTATATE PET/CT imaging. Next, the variable acquisition time regimen was validated and assessed for agreement with objective and subjective image quality evaluations in a validation cohort. The method we designed for a variable acquisition regimen might be instructive for other PET systems. Image quality is commonly evaluated by the metric SNR or noise equivalent count rate (NECR). The SNR is calculated as the value of the square root of the product of system sensitivity, injected activity, and acquisition time [13], whereas the NECR is calculated as the ratio of the square of the true events to the total of true events, random events, and scatter coincidences [18]. A constant value of SNR and NECR can overcome the limitation of patient-specific attenuation to render uniform image quality. To achieve a constant image, acquisition time should vary from different patient size. For example, a typical patient with 100 kg should scan more time than a patient with 50 kg to achieve target SNR. One patient scanning by a related lower sensitivity detector should prolong acquiring time than by higher sensitive detector. In addition, to achieve target SNR, low activity of tracer should have longer scanning time than high activity. In developing cohort, the SNR L increased with increasing acquisition time within 1 min (Fig. 2a). The SNR norm showed a strong correlation with body mass (less than 150 kg) with a determination coefficient (R square) of 0.63, slightly lower than that in a previous study [13]. We speculate that the range of body mass and the sample size might have contributed to this difference.
Based on the excellent intra-and inter-agreement agreement (all kappa > 0.85), the mean threshold SNR L was 11.2 for acceptable image quality of 68 Ga-DOTA-TATE total-body PET/CT. Compared with 18 F-FDG images we previously analyzed, the threshold SNR of 14.0 was slightly higher. The difference was consistent with a previous study in which the acceptable SNR L was 6.2 for whole-body 68 Ga-DOTATATE PET/CT, but 18 F-FDG studies have revealed a higher SNR of 10 [19]. Compared with 18 F-FDG with liver SUVmean of 2.6, the accumulation degree of 68 Ga-DOTATATE in liver was higher with liver SUVmean of 8.4 [11]. We hypothesize that difference might be related to the percentage of biological distribution in the liver. In addition, the coefficient of variation, representing image noise, is recommended to be 15% as a reference maximum noise level for clinical 18 F-FDG PET image interpretation [20]. In the present study, both the CV of D2 in the development cohort and R4 in the validation cohort were less than 10%. A recent study [9] showed that 5.5 times noise reduction of a 194-cm FOV PET compared to a 30-cm FOV digital PET with the same total examination time for scanning a 2-m-long phantom, and the noise reduction became 1.5 times when the same acquisition time per bed was performed. In the validation cohort, all 90 lesions were detected in all acquisition time PET images of all subgroups. The higher percentage of G1 and G2 patients with marked SSTR expression (18/19 for the development cohort and 47/57 for the validation cohort) than G3 or NEC patients enrolled in this study might lead to bias in the results. Based on the proposed variable time regimen, the mean time was 2.99 ± 0.91 min, ranging from 2.18 to 6.35 min. Compared to the fixed acquisition protocol of 10 min, the mean acquisition time decreased by 70.1%, ranging from 36.5% to 78.2%. The variable acquisition time regimen based on a constant SNR L of 11.2 of total-body scan can eliminated the influence of BM, and provide more consistent image quality.
In this study, SNR was selected for evaluation of image quality because there was relatively homogeneous uptake of 68 Ga-DOTATATE by the liver, which was easily influenced by several circumstances. Previous studies have found less uptake of 68 Ga-DOTATATE by the liver, spleen, and thyroid after treatment initiation in patients with than without somatostatin analog treatment [21]. One prospective study performed 68 Ga-DOTATATE imaging one day before and one day after injection of lanreotide, and no evidence of decreased uptake in the tumor, but a higher tumor-to-liver ratio, was obtained [22]. To avoid the influence of treatment, our study excluded 6 patients with continuous treatment with octreotide and one patient with high-intensity focused ultrasound in the liver within 1 week. Additionally, the dynamic distribution between the development cohort regimen and low-activity regimen exhibited equivalent trends for the liver, pancreas, kidney, and spleen over time, which might eliminate the influence of different doses on biological distribution (Fig. S2). In addition, the image quality could also be improved using the reconstructed method of PSF and TOF. Previous study reported that the TOF could obtain more contract information than that without TOF information [23]. Although the image reconstructed with PSF correction slowed the iterative convergence, it could provide a more uniform background and increased SNR than that without PSF correction. Previous study showed that the sufficient image quality could be acquired for low activity objects and a shorter acquisition time when the image constructed by PSF and TOF [24]. In this study, the sufficient image quality might be contributed by the combination of ultra-high sensitivity of total body detector and the image reconstructed by TOF and PSF.
The findings of this study have to be considered in light of several limitations. First, the variable acquisition time regimen was established based on retrospective image reconstructions for 19 full-activity cases. The small sample size may result in confounding bias that may influence the reliability of the proposed regimen. Second, further investigation focusing on more organ SNRs, such as the spleen, kidney and additional metrics for image quality are needed. Third, we provide a method to realize personalized duration time; however, the acquisition time regimen was established only on total body PET. Thus, the proposed method should be referred to and rebuilt for other PET systems according to those characteristics.

Conclusion
A BM-specific acquisition time regimen was proposed and validated in patients with NETs on 68 Ga-DOTA-TATE total-body PET/CT imaging. Based on the proposed regimen, the homogenous image quality with reasonable acquisition time was available for a constant level, independent of body mass.