Can the BMI-based dose regimen be used to reduce injection activity and to obtain a constant image quality in oncological patients by 18F-FDG total-body PET/CT imaging?

PET image quality is influenced by the patient size according to the current guideline. The study aimed to propose an optimized dose regimen to yield a constant image quality independent of patient habitus to meet the clinical needs. A first patient cohort of 78 consecutive oncological patients (59.7 ± 13.7 years) who underwent a total-body PET/CT scan were retrospectively enrolled to develop the regimen. The patients were randomly distributed in four body mass index (BMI) groups according to the World Health Organization (WHO) criteria. The liver SNR (signal-to-noise ratio, SNRL) was obtained by manually drawing regions of interest (ROIs) and normalized (SNRnorm) by the product of injected activity and acquisition time. Fits of SNRnorm against different patient-dependent parameters were performed to determine the best correlating parameter and fit method. A qualitative assessment on image quality was performed using a 5-point Likert scale to determine the acceptable threshold of SNRL. Thus, an optimized regimen was proposed and validated by a second patient cohort consisted of prospectively enrolled 38 oncological patients. The linear fit showed SNRnorm had the strongest correlation (R2 = 0.69) with the BMI than other patient-dependent parameters and fit method. The qualitative assessment indicated a SNRL value of 14.0 as an acceptable threshold to achieve sufficient image quality. The optimized dose regimen was determined as a quadratic relation with BMI: injected activity (MBq) = 39.2 (MBq)/(− 0.03*BMI + 1.49)2. In the validation study, the SNRL no longer decreased with the increase of BMI. There was no significant difference of the image quality regarding the value of SNRL between different BMI groups (p > 0.05). In addition, the injected activity was reduced by 75.6 ± 2.9%, 72.1 ± 4.0%, 67.1 ± 4.4%, and 64.8 ± 3.5% compared with the first cohort for the four BMI groups, respectively. The study proposed a quadratic relation between the 18F-FDG injected activity and the patient’s BMI for total-body 18F-FDG PET imaging. In this regimen, the image quality can maintain in a constant level independent of patient habitus and meet the clinical requirement with a reduced injected activity.

F-18-fluorodeoxyglucose positron emission tomography/ computed Tomography ( 18 F-FDG PET/CT) has been widely used in tumor diagnosis, staging, restaging, and response evaluation. Meanwhile, this hybrid imaging modality has shown the potential value in prognosis prediction and imageguided biopsy, providing both anatomic and functional information for clinical management [1][2][3][4][5]. The total-body PET/ CT, uEXPLORER (United Imaging Healthcare, China), with an increased geometric coverage to encompass the entire body, can dramatically improve the PET sensitivity by a factor of about 40 over existing PET scanners for imaging the entire body. This predicted gain in sensitivity has various implications, such as to improve the image quality to reconstruct images with higher resolution and allow detection of smaller or lower-contrast structures. In addition, it can be used in the clinical practice with the reduced injected activity or short PET acquisition duration while maintaining the image quality [6].
Previously, our team has conducted a series of research on PET image quality with the uEXPLORER. Zhang et. al [7] found that oncological patients with an injected activity of 4.4 MBq/kg and an acquisition time of 30-60 s could obtain an acceptable image quality for total-body PET imaging. Furthermore, our previous results showed that total-body PET/CT with half-dose (1.85 MBq/kg) 18 F-FDG of 2-min acquisition could achieve an equivalent image quality to that of whole-body PET/CT with full-dose (3.7 MBq/kg) in lung cancer [8]. Recently, Liu et.al reported that dynamic PET imaging of ultra-low-dose (0.37 MBq/kg) injected activity achieved relevant kinetic metrics of 18 F-FDG and comparable image contrast with full-dose imaging [9].
However, these studies have all adopted a weight-based injected activity as recommended in the European Association of Nuclear Medicine (EANM) Guideline [10]. It is well-known that the linear weight-based dose regimen had several deficiencies for a stable and reliable PET image quality, especially in obese patients with a significantly high amount of body fat, resulting in low FDG accumulation [11]. Thus, this study aimed to further investigate the influence of patient habitus on PET image quality and propose a personalized dose regimen to yield a more constant image quality.

Patients
The study included two cohorts to develop and validate the new dose regimen for total-body PET imaging. We retrospectively enrolled 78 consecutive patients as the first patient cohort, who were referred to our center for 18 F-FDG PET/CT examinations from September 2019 to July 2020. They were randomly selected from the database of our center. Patients with severe fatty liver, cirrhosis, and multiple liver metastasis were excluded. The included patients were equally distributed in each BMI group according to the criteria of the WHO [12], with 20 patients in the underweight group, 19 patients in the normal-weight group, 19 patients in the overweight group, and 20 patients in the obese group. Subsequently, the second patient cohort consisting of 38 patients with known or suspected malignancy were prospectively enrolled to validate the proposed dose regimen, including 7 patients, 10 patients, 11 patients, and 10 patients for each BMI group, respectively. Exclusion criteria for the second patient cohort are those with diabetes or younger than 18 years old. The demographic characteristics of patients in the two cohorts were extracted from the database, including gender, age, body mass (BM), and height (H). BMI (kg/m 2 ) was calculated by dividing the BM (kg) by the square of height (m). Considering human body composition, lean body weight (LBW, kg) and fat mass (FM, kg) were also investigated as the patient-dependent parameters in the study. LBW for male and female as well as FM was calculated as follows (Eqs. 1-3, respectively) [13].
In Eqs. 1 and 2, LBW is the lean body weight in kilograms, H is the height in meters, BM is the body weight in kilograms, and age is patient age in years. In Eq. 3, FM is the fat mass in kilograms, BM is the body weight in kilograms, and LBW is the lean body weight in kilograms.
This study was approved by the Institutional Review Board of Zhongshan Hospital, Fudan University. Informed consent was waived to the patients in the first cohort due to the retrospective nature and all patients in the second cohort signed an informed consent prior to the PET/CT scan.

PET/CT examination
All patients were instructed to fast and avoid strenuous exercise at least 6 h prior to the 18 F-FDG injection, and blood glucose level was measured and recorded. In the study on the first patient cohort, a bolus injection of 18 F-FDG (3.7 MBq/ kg) was intravenously administered. In the study of the second patient cohort, the injection activity was strictly following the proposed dose regimen. All images were acquired on the uEXPLORER. A CT scan was performed before PET imaging for attenuation correction and anatomical localization with a dose modulation technique. Subsequently, a totalbody PET imaging was performed with 5-min acquisition with arms down positioning. (1) PET raw data was segmented into 30, 45, 60, and 120 s from the 300 s list-mode data, referred as G30, G45, G60, G120, and G300. All the PET images were reconstructed using a 3D ordered subset expectation maximization algorithm with the following parameters: 3 iterations, 20 subsets, a matrix of 192 × 192, slice thickness of 1.443 mm, time of flight and point spread function modeling. A Gaussian filter with a full width at half maximum of 3 mm was applied to the reconstructed images.

Image analysis
In PET clinical studies, the signal-to-noise ratio in the liver (SNR L ) was used as a measure of image quality as it is the organ with a relatively homogeneous uptake of FDG in the human body. It is well-known that various factors, such as the patient weight, injected activity, and acquisition time, can impact the SNR L .
For a given situation on a PET scanner, SNR in PET images is dominated by the Poisson statistics inherent in radionuclide decay detection and is proportional to the square root of the detected events. In the first-order approximation we expect that [7].
where k is a constant, S is the effective sensitivity of the scanner, A is the injected activity (MBq), and t is the acquisition time per bed position (min). Here, the dose-time product (DTP, MBq·min) is the product of the injected activity (MBq) and the acquisition time per bed position (min). If the SNR in the liver is normalized by the square root of the DTP, it can be assumed to be independent of the injected activity and the acquisition time (Eq. 5) [14]. Therefore, SNR norm (1/sqrt (MBq·min)) can be regarded as a function of patient-dependent parameters and investigated in the study.
In the first part of the study, the slice with the largest cross section in the liver on the CT transverse slice was determined. In the corresponding PET slice and two adjacent slices, a circular region of interest (ROI) with a diameter of 20 ± 1 mm was manually drawn in a lesion-free and homogenous region of the right liver lobe with care to avoid large blood vessels and the partial volume effect ( Supplementary  Fig. 1). The ROIs were identical in all the three slices. Liver standard uptake value (SUV) and its standard deviation (SD) were measured and recorded and determined as the average of three ROIs. The SNR L was obtained by dividing the liver SUV mean by its SD (Eq. 6) [15].
Both linear and non-linear fits were performed with the SNR norm vs. the patient-dependent parameters. The highest coefficient of determination (R 2 ) was used to determine the best-correlated parameter and fit method. In addition, the relation between the SNR norm and the patient-dependent parameters, referred as SNR fit were obtained from the fit function (Eq. 7).
where p indicated the best-correlated parameter and a, b, and c were constants derived from the fit function.
In order to determine the acceptable SNR L threshold (SNR acc ), a qualitative analysis on image quality was performed. The image quality was independently assessed by two experienced nuclear medicine physicians on a dedicated workstation (uWS, United Imaging Healthcare, Shanghai, China). For each patient, the reading order of PET images was randomized by an independent operator. The patient's history and the acquisition time were blinded to the readers. Image quality was assessed with a 5-point Likert scale (1 = non-diagnostic image quality; 2 = poor image quality; 3 = moderate image quality; 4 = good image quality; 5 = excellent image quality). Figure 1 shows reference images with different Likert scores. The score of 3 was equivalent to the image quality to meet the clinical need in our department and served as the reference to determine the acceptable threshold (SNR acc ). The SNR acc was obtained by calculating the mean value of SNR L from all the images scored with 3 points. Finally, the dose regimen was determined as follows (Eq. 8).
where t indicated the acquisition duration and SNR fit was the determined function of the fit to the SNR norm vs. patientdependent parameter.
Subsequently, the proposed dose regimen was validated with a newly enrolled patient cohort. In the validation group, the patient was injected strictly following the proposed regimen. Image quality was assessed qualitatively using the same criteria by the same nuclear medicine physicians. Quantitative analysis was performed to compare the liver SUV mean , SD, as well as SNR L between the two cohorts. For each patient, the lesion with the highest uptake was selected for analysis. A volume of interest (VOI) was manually drawn on each selected lesion and SUV max was obtained and compared between the two cohorts.

Statistical analysis
All statistical analysis was performed using SPSS Statistics Version 26 (IBM Inc., Chicago, IL, USA) and GraphPad Prism 8 (GraphPad Software Inc., San Diego, California, USA). Data were described as mean ± SD. Differences in quantitative variables were assessed by analysis of variance (ANOVA) with post hoc Bonferroni adjustment for pairwise comparison. Independent t test was performed to compare the quantitative variables between the two cohorts. Categorical variables were compared using the Chi-square test. Cohen's kappa analysis was performed to evaluate the inter-reader agreement. Wilcoxon signed rank test was used to compare the qualitative scores between different BMI groups. Statistical significance was considered if p value is less than 0.05.

Patient characteristics
The demographic and clinical characteristics of the two patient cohorts are listed in

The development of the dose regimen
The SNR L increased along with the increase of acquisition duration with a significant difference to that in G300 (as shown Fig. 2). Moreover, SNR L showed significant difference between BMI groups (all p < 0.01), as listed in Table 2.
The SNR L decreased along with the increase of the BMI groups for a given group, as observed in the clinical practice. Compared with that in the normal weight group, the SNR L in other BMI groups were significantly different (all p < 0.05). As expected, the SNR L of the obese group had the lowest value, indicating the worst image quality. The SNR norm , the normalized SNR L , was fitted with the different patient-dependent parameters using a linear and non-linear fit method, as illustrated in Fig. 3. It was found that the SNR norm was best fitted with BMI with a linear fit function, with the highest coefficient (R 2 = 0.69) and slightly lower coefficient in a non-linear fit with BMI (R 2 = 0.68). Therefore, BMI was determined as the best-correlated parameter, with a linear fit function (Eq. 9). The image quality of G300 was used as a reference in the study and scored with 5 points from both readers. The overall image quality of G30, G45, G60, and G120 were scored with 2.09 ± 0.29, 2.47 ± 0.50, 2.97 ± 0.48, and 3.82 ± 0.38, respectively, as presented in Table 3. The inter-reader agreement of overall image quality showed a substantial result with a kappa value of 0.93, 0.99, 0.95, and 0.94 for each acquisition duration, respectively. Compared with the reference, there was a significant difference regarding the qualitative scores in G30, G45, G60, and G120 (all p < 0.05). A SNR acc value of 14.0 was obtained by calculating the average of the SNR L of all the images with a score of 3 points. For a given acquisition duration of 5 min, the value of SNRacc^2/t was derived as 39.2. Thus, the new dose regimen can be determined as (Eq. 10):

Validation of the proposed regimen
The new dose regimen was validated with a second patient cohort consisting of 38 oncological patients both   According to Whiskers specify Tukey plot, a significant difference compared with G300 (grey box) is indicated with **** (p < 0.0001) qualitatively and quantitatively. In the qualitative analysis, there was no significant difference between the BMI groups (as shown in Fig. 4), indicating a constant image quality. Compared with that in the weight-based regimen, the proposed regimen can improve the image quality of the patients with a BMI no less than 25, as shown in Fig. 5.
In the quantitative analysis, there were no significant differences in the liver SUV mean and lesion SUV max between the patient cohorts (p > 0.05, as shown in Table 4). The liver SD in the second patient cohort showed a significantly larger value than that in the first patient cohort due to the reduced injected activity. The SNR L was plotted vs. the BMI, as  Fig. 3 The linear fits (a-g) and non-linear fits (a1-g1) of the SNRnorm against the patient-dependent parameters, including height (a, a1), BM (body mass, b, b1), body mass per height (BM/H, c, c1), lean body weight (LBW, d, d1), fat mass (FM, e, e1), body surface area (BSA, f, f1) and body mass index (BMI, g, g1). Note: SNRnorm, normalized signal-to-noise ratio shown in Fig. 6. The SNR L in the validation cohort showed no decreased tendency with the increase of BMI, indicating a constant image quality independent of patient habitus. In addition, there was no significant difference of the SNR L between different BMI groups, as shown in Fig. 7. Furthermore, the injected activity of the two patient cohorts was investigated. There was significant difference between the two cohorts regarding the injection activity (236. 8  Compared with the first patient cohort, the reduction of the injected activity in the second cohort was up to 69.2 ± 5.4%.

Discussion
The current EANM guideline recommended a linear weightbased regimen for 18 F-FDG PET examinations [16]. A quadric relationship between the 18 F-FDG administered activity, PET acquisition time, and patient BM was also described in the guideline [10]. In this regimen, the SNR L for the patient with a BM ≥ 75 kg will be decreased, indicating a degraded image quality due to excessive attenuation and scatter. Thus, an experienced technician is required to modify the acquisition scheme in the clinical scenario for specific situations, which inevitably complicates the operations. Previous studies utilized a higher 18 F-FDG activity per kilogram for patients with a body mass ≥ 90 kg to compensate for attenuation, while SNR L still decreased with body mass in the overweight group [15]. In previous studies, SNR L of 9.6 and 10.0 could yield a good image quality [15,17]. Tan et al. reported that the SNR L of 11.7 in the half-dose total-body group with a 2-min duration was higher than the SNR L of 8.3 in fulldose whole-body group. However, in this study, the subjective assessment of image quality found that a SNR L of 14.0 could obtain a sufficient image quality. A reason for the higher SNR L in this study may be bias from different raters. A quadratic relation between the BMI and injected 18 F-FDG activities was determined in this study, which contributes to more constant image quality not affected by BMI. Although this regimen is less convenient than a linear relation in clinical practice, it can be easily overcome by an automatic calculator or a look-up table.
It has been known that the patient-dependent parameters can influence the PET image quality, which is the initial motivation of this study. The fits of SNR norm with different parameters showed that both the quadric and linear fitting with BMI had the highest R 2 (0.68 and 0.69, respectively). Other parameters, such as the BM, height, body mass per height, LBW, FM, and BSA showed a lower value of R 2 than the parameters discussed in previous studies [14,15]. Based on the results, a linear fit with BMI was selected. The findings were inconsistent with previous studies which suggested a quadratic dose regimen of BM. This may be due to the difference of the subjects selected in the studies. The BM range and number of the enrolled patients may impact the results. Moreover, the population body shape varies with races, which might also induce a bias to the results.
In the proposed regimen, the injected activity was with a 69.2 ± 5.4% reduction compared with that in the weightbased regimen (3.7 MBq/kg). In addition, the injected activity was reduced by 75.6 ± 2.9%, 72.1 ± 4.0%, 67.1 ± 4.4%, and 64.8 ± 3.5% for the underweight, normal weight,  [8]. It is due to the high sensitivity of the total-body PET scanner which is about 40-fold of that for a conventional PET scanner [10]. This helps to propose the regimen with a reduced injected activity while maintaining the image quality feasible for clinical practice. Obviously, the proposed regimen had limited application in patients due to the methodology. The patient with a BMI ≥ 35 was not enrolled in the study due to the limited patient weight in our site. Actually, the obese patient referred a PET/CT scan in our center were scarce. Therefore, extrapolation was simply used to develop the regimen. Due to the mathematical nature of the quadric expression, the injected activity was dramatically increased with the increase of BMI values. Therefore, an upper limit should be determined for safety concern. Here, we simply investigated the SNR L against BMI in a linear relationship: SNR L = − 0.5*BMI + 33.1. Since the acceptable of the SNR L should be more than 14.0 to meet the need of image quality, the upper limit of BMI was determined as 38.2 kg/ m 2 . Thus, the proposed regimen may be not feasible for the patient with a BMI larger than the upper limit. In this study, SNR measured in the liver was selected as a measure to assess the image quality since the liver has a relatively homogeneous uptake of 18 F-FDG. However, SNR could be influenced by several physiological factors, such as blood glucose levels, uptake time, plasma clearance, and drinking water status. Blood glucose Fig. 5 Comparison of patient images in a linear weightbased (a, b) and the proposed BMI-based dose regimen (c, d). Compared with the linear weight-based regimen, MIP and transverse images of the total-body 18 F-FDG PET images showed an improved image quality for patients with a BMI = 25 kg/m 2 (subfigure a vs. c) and 30 kg/m 2 (subfigure b vs. d) Table 4 Comparison of Liver SUVmean, SD and lesion SUVmean between the patient cohorts Independent t test was used to compare the difference between the two patient cohorts, and an asterisk indicated a significant difference. SUV, standardized uptake value Data were presented as mean ± standard deviation levels affects liver uptake of the activity as reported in other studies [18]. To minimize the influence factor, the glucose level was controlled within a normal range (3.9-6.1 mmol/L) for both cohorts. Further study should be performed with the glucose level in a wider range. Additionally, plasma clearance can be influenced by water consumption before the scan and the distribution of FDG may be changed. According to our experience, although the patients were recommended to drink 0.5-1L of water after the injection of FDG, not all the enrolled patients strictly followed this instruction. The uptake time of FDG also influenced the level of plasma clearance and the liver SNR. A previous study reported that liver SUV remains constant if the uptake time is in the range of 50-110 min [19]. The uptake time varied from 45 to 121 min in the study, and the results may be biased. The study has several limitations. Firstly, both cohorts do not include subjects with a BMI larger than 35. Secondly, it is a single-center preliminarily study, and the number of patients in the validation cohort was limited. The proposed regimen should be further validated in a multi-center large scale study. Thirdly, this study just simplified the regimen as a function of the injected activity. In future study, the combination of personalized acquisition time and injected activity should be further explored.

Conclusion
The study recommended a quadratic relation between the 18 F-FDG injected activity and the patient's BMI and proposes a regimen for total-body PET imaging. In the regimen, the image quality can maintain in a constant level independent of patient habitus and meet the clinical requirement even with a reduced injected activity.
Authors' contributions Jie Xiao was involved in the statistical analysis and manuscript writing. Haojun Yu contributed to data acquisition and reconstructions. Xiuli Sui and Yan Hu contributed to data analyses and image interpretation. Guobing Liu and Yanyan Cao helped with data processing. Yiqiu Zhang and Pengcheng Hu supervised the study. Ying Wang and Chenwei Li were with editing English grammar. Hongcheng Shi and Baixuan Xu designed the study and contributed to editing and reviewing the manuscript. All authors read and approved the final manuscript.

Declarations
Ethical approval All procedures performed in studies involving human participants were in accordance with the ethical standards of the insti- Fig. 6 Scatter plots of the SNR L vs. BMI in the two patient cohorts. In the first patient cohort, the SNR L showed a decreased tendency along with the increase of BMI, whereas there was no such tendency in the validation cohort Fig. 7 The comparison of SNR L in different BMI groups in the two cohorts. ***, significant difference (p < 0.001); ns, no significant difference