Selection of SUV normalization and reference tissue for quantitative [68Ga]Ga-Pentixafor PET/CT analysis

Abstract

Background

[68Ga]Ga-Pentixafor PET/CT enables visualization of C-X-C chemokine receptor 4 expression, but quantitative accuracy depends on standardized uptake value (SUV) normalization and reference tissue selection. Body composition and variability across candidate organs may confound SUV interpretation.We retrospectively analyzed [68Ga]Ga-Pentixafor PET/CT scans from 40 patients without malignancy. Regions of interest were placed in the blood pool, liver, spleen, bone marrow, and muscle. Three SUV were calculated: SUV normalized by body weight (SUVbw) and SUV normalized by lean body mass (SUL), the latter derived using two methods—the conventional James formula and the Janmahasatian (Janma) formula, a revised method optimized for obese individuals. Correlations with body weight were quantified using Spearman correlation coefficients and the slope-to-intercept ratio (a/b) from linear regression analyses, while interobserver reproducibility and variability were evaluated using the intraclass correlation coefficient (ICC) and the coefficient of quartile variation (CQV), respectively. This study aimed to identify an optimal normalization strategy and reliable reference tissue for quantitative [68Ga]Ga-Pentixafor PET/CT.

Results

SUVmean, SULmean-James, and SULmean-Janma exhibited significant body-weight dependence in all candidate tissues except muscle, with the strongest association observed in the liver (right liver lobe: ρ = 0.635, 0.576, and 0.650, respectively). Compared with SUVbw, SUL derived from the James formula effectively reduced body-weight-related fluctuations in SUV measurements: a/b decreased from 0.065 to 0.034 in the right liver lobe, 0.036 to 0.021 in the left liver lobe, and 0.076 to 0.037 in the spleen. Across all normalization methods, SUV measurements demonstrated good repeatability in all candidate reference tissues (ICC > 0.7). Variability analysis showed that the liver and blood pool yielded the lowest dispersion for both SUVmean and SULmean-James (CQV: 9.4%-16.8%), identifying them as the most stable reference tissues. In contrast, splenic uptake exhibited substantial inter-individual variability (CQV: 19.9%-22.2%), suggesting it is unsuitable as a routine reference tissue.

Conclusion

SUL-James normalization, together with the liver or blood pool as reference tissues, provides robust and reproducible quantification for [⁶⁸Ga]Ga-Pentixafor PET/CT. These findings support standardized imaging frameworks and may facilitate treatment response evaluation, prognostic assessment, and harmonization across centers.

Introduction

C-X-C chemokine receptor 4 (CXCR4) is a G protein-coupled receptor for stromal cell-derived factor-1 (SDF-1) [1]. The SDF-1/CXCR4 axis plays a pivotal role in cell migration, homing, and inflammatory responses. In tumors, this pathway is often aberrantly activated, thereby promoting proliferation, invasion, and metastasis [2]. [⁶⁸Ga]Ga-Pentixafor is currently the most widely used CXCR4-targeted PET tracer, binding specifically to CXCR4-positive cells and enabling visualization of this pathway. Its clinical utility has been demonstrated not only in hematologic malignancies and solid tumors, but also in various non-neoplastic conditions including endocrine and inflammatory diseases [3–5].

The standardized uptake value (SUV) is the most used semiquantitative parameter in PET/CT. By comparing tracer uptake in lesions with reference tissues, SUV improves the objectivity and consistency of visual interpretation. However, SUV calculation is influenced by multiple factors, among which the choice of normalization method exerts a significant impact on SUV absolute values [6–9]. Traditionally, SUV is normalized to body weight (BW) (also written as SUVbw). For tracers with low uptake in adipose tissue, SUVbw may be substantially affected by variations in body fat, making measurements less reliable in patients with abnormal body composition, such as children or those with cachexia. To overcome this limitation, lean body mass (LBM) normalization has been introduced, yielding the LBM-corrected SUV (SUL). Notably, SUL has been recommended by the PERCIST criteria for treatment response assessment [10], particularly in populations with significant weight fluctuations. A similar issue may also arise in [⁶⁸Ga]Ga-Pentixafor PET, where differences in body composition could confound SUV accuracy.

The choice of reference tissue is critical for both image interpretation and quantitative analysis. For example, in [¹⁸F]F-FDG PET, the Deauville score enhances interpretative consistency by comparing lesion uptake with that of the liver or blood pool [11]. In contrast, [⁶⁸Ga]Ga-Pentixafor has distinct uptake characteristics, and there is no consensus on the most appropriate reference tissue. Although a few studies have described its biodistribution and dynamic imaging, systematic evaluation of the stability and reliability of candidate reference tissues remains lacking [12–15].

Based on this background, the present study aimed to systematically investigate SUV normalization methods and reference tissue selection for [⁶⁸Ga]Ga-Pentixafor PET/CT: (1) to compare BW- versus LBM-based normalization and assess their ability to reduce SUV dependence on body weight; and (2) to identify reference tissues with low variability and high reproducibility, thereby improving the accuracy and comparability of quantitative assessment in both clinical practice and research settings.

Materials and methods

Patients

We retrospectively collected patients diagnosed with primary aldosteronism who underwent [⁶⁸Ga]Ga-Pentixafor PET/CT at our institution between January 1 and December 31, 2023. A total of 40 patients, including 20 males and 20 females, were randomly selected from this cohort and analyzed as the healthy control cohort. The inclusion criteria were: (1) age > 18 years; (2) no history of malignancy; (3) absence of malignant disease during at least one year of follow-up after imaging. The exclusion criteria were: (1) autoimmune disorders; (2) impaired hepatic or renal function; (3) major illness or severe trauma within the past three months; and (4) incomplete clinical data. This study was approved by the institutional ethics committee (Approval No. KY2025316).

Preparation and imaging of [⁶⁸Ga]Ga-Pentixafor

The pentixafor precursor purchased from Invivo Chemical Technology Co., Ltd. (Guangzhou, China) was dissolved in deionized water to prepare a 1 mg/mL solution, which was stored at 4 °C for later use. Subsequently, the solution containing [⁶⁸Ga]Ga was obtained by eluting the [⁶⁸Ge]Ge/[⁶⁸Ga]Ga generator (China Isotope and Radiation Corporation, Beijing) with 4 mL of 0.1 mol/L hydrochloric acid. Mix 1 mL of 1.25 mol/L sodium acetate buffer with the above [⁶⁸Ga]Ga solution, transfer it to an EP tube containing 20 µg of pentixafor, and react at 95 °C for 15 min to complete the labeling. After the reaction, the solution was cooled to room temperature, purified using a C18 solid-phase extraction column, eluted with 75% ethanol, and filtered through a 0.22 μm sterile filter membrane to obtain the final product. The final [⁶⁸Ga]Ga-pentixafor product was analyzed by radio-HPLC, with a radiochemical purity exceeding 95%.

Patients require no special preparation prior to [⁶⁸Ga]Ga-Pentixafor PET/CT imaging. The [⁶⁸Ga]Ga-Pentixafor injection dose is calculated based on patient weight (0.05 mCi/kg). Forty to sixty minutes after intravenous injection of the radiotracer, patients undergo PET/CT scanning (uMI 780, United Imaging Healthcare, Shanghai, China). The scan begins with CT acquisition, covering the region from the top of the head to the upper thigh, with a slice thickness of 3 mm, a tube voltage of 120 kV, and tube current automatically adjusted by the device based on radiation attenuation. The resulting CT images are used for anatomical localization and attenuation correction of PET images. PET acquisition is then performed on the same bed as the CT scan, using a three-dimensional mode, with each bed acquisition lasting 3 min. All raw data were transferred to the post-processing workstation (uWS-MI, version R002, United Imaging Healthcare) and reconstructed using the ordered subset expectation maximization (OSEM) algorithm with reconstruction parameters of 2 iterations and 20 subsets.

SUV measurement

Two experienced nuclear medicine physicians manually delineated regions of interest (ROIs) on each patient’s [⁶⁸Ga]Ga-Pentixafor PET/CT images. ROIs were defined as spherical regions with a diameter of 2 cm, located in the blood pool (descending aorta), right liver lobe, left liver lobe, spleen, bone marrow (L3 vertebra), and muscle (L3 vertebral level). SUVmean (for BW-normalized SUV) and SULmean (for LBM-normalized SUV) for each ROI were recorded. Fifteen days after the initial segmentation, the two physicians independently re-segmented the same set of images. The final SUV measurements were calculated as the average of the four segmentation results. LBM is calculated by James [16] and Janma’s [17] method.

Statistical analysis

Data following a normal distribution are presented as mean ± standard deviation (SD). Non-normally distributed data are expressed as median (IQR), and their dispersion was assessed using the coefficient of quartile variation (CQV). CQV is calculated as the percentage ratio of IQR to the median and is more suitable than the coefficient of variation for skewed data; lower CQV values indicate smaller variability. Measurement consistency was assessed using the intraclass correlation coefficient (ICC). Correlations were analyzed using Spearman rank correlation, with the correlation coefficient (ρ) used to quantify the strength of associations between variables. Simple linear regression models were subsequently constructed to quantify the effect of body weight on each dependent variable. The F-test was used to assess the overall significance of the model. The coefficient of determination (R²) reflected the proportion of variance in the dependent variable explained by the independent variable. The ratio of the slope (a) to the intercept (b) was used to describe the relative percentage change in SUV or SUL associated with changes in body weight. This ratio normalizes the slope by the intercept to eliminate baseline differences between models and thereby allows a more accurate comparison of the degree of weight dependence across models, functioning similarly to a coefficient of variation. P < 0.05 was considered statistically significant. All statistical analyses and graphical representations were performed using Python (version 3.7.16).

Results

Patient characteristics

This study included [⁶⁸Ga]Ga-pentixafor PET/CT images from 40 patients (20 males and 20 females) for analysis. Overall, the patients had an average weight of 65.1 ± 11.2 kg, and an average height of 161.6 ± 7.4 cm. The mean LBM estimated using the James formula and Janma formula were 48.2 ± 8.1 kg and 45.5 ± 9.7 kg, respectively.

SUV normalization based on BW and LBM methods

Figure 1 shows an example of ROI delineation in a [⁶⁸Ga]Ga-Pentixafor PET/CT image of a patient.

Fig. 1.

Example of Region-of-Interest delineation on ⁶⁸Ga-Pentixafor PET/CT for reference tissues. (The regions are color-coded as follows: red, blood pool; yellow, right hepatic lobe; blue, left hepatic lobe; purple, spleen; pink, bone marrow; green, muscle.)

The correlation between body weight and SUV/SUL varied across different tissues depending on the normalization method. Using SUVmean, SULmean-James, and SULmean-Janma as indicators, we found that values in the right liver lobe (ρ = 0.635, 0.576, 0.650, respectively), left liver lobe (ρ = 0.599, 0.511, 0.609, respectively), spleen (ρ = 0.395, 0.345, 0.468, respectively), and bone marrow (ρ = 0.450, 0.433, 0.492, respectively) were positively correlated with body weight under all normalization methods (p < 0.001 to p = 0.029). SUVmean (ρ = 0.373, p = 0.018) and SULmean-Janma (ρ = 0.438, p = 0.005) of the blood pool were also positively correlated with body weight, whereas no significant correlation was observed for SULmean-James (p = 0.073). In contrast, muscle SUVmean showed no significant correlation with body weight under any normalization method (p = 0.075–0.304). Overall, among all candidate tissues except muscle, SULmean-James consistently exhibited the lowest dependence on body weight. Notably, after LBM correction using the Janma formula, the correlation between SUL and body weight in some tissues was even stronger than that observed with BW correction.

We then performed linear fitting between SUV/SUL and body weight to more precisely quantify the magnitude of change. Because muscle SUVmean, SULmean-James, SULmean-Janma, and blood pool SULmean-James showed no significant correlations with body weight, these metrics were excluded from regression analysis. In addition, SULmean-James of the bone marrow did not yield a significant linear regression model (p = 0.0502) and was therefore excluded as well. Consistent with the correlation analysis, SULmean-James showed weaker weight dependence than SUVmean in the right liver lobe (a/b: 0.034 vs. 0.065), left liver lobe (a/b: 0.021 vs. 0.036), and spleen (a/b: 0.037 vs. 0.076). Detailed results and trends are presented in Fig. 2.

Fig. 2.

Scatter plots and linear regression of mean Standardized Uptake Value (SUVmean/SULmean) in background organs based on Body Weight (BW) and Lean Body Mass (LBM) Correction. (A) Left and right hepatic lobes and muscle; (B) blood pool and bone marrow; (C) spleen. ρ and p represent the Spearman correlation coefficient and its significance level, respectively. For indicators with significant correlations, linear regression was further performed, where p indicates the statistical significance of the regression line (a dashed line denotes non-significance). R² is the coefficient of determination. The slope (a) and intercept (b) ratio quantifies the relative dependence of SUVmean/SULmean on body weight, minimizing the impact of intercept differences on slope interpretation

Variability of SUV measurements across reference organs

In the repeated-measurement analysis, SUVmean, SULmean-James, and SULmean-Janma for the blood pool (ICC = 0.93, 0.92, 0.94, respectively), right liver lobe (ICC = 0.95, 0.98, 0.98, respectively), left liver lobe (ICC = 0.86, 0.86, 0.89, respectively), spleen (ICC = 0.98, 0.92, 0.93, respectively), bone marrow (ICC = 0.92, 0.95, 0.96, respectively), and muscle (ICC = 0.78, 0.75, 0.78, respectively) all demonstrated high repeatability, with ICC values exceeding 0.7 across all normalization methods. Two-way repeated-measures ANOVA further revealed significant inter-observer bias in SUVmean, SULmean-James, and SULmean-Janma of the blood pool (p = 0.03, 0.03, 0.04, respectively) and muscle (p = 0.03 for all), as well as in splenic SUVmean (p = 0.04) (Table 1).

Table 1.

Consistency of standardized uptake value measurements in repeated assessments

Organs	Method	ICC	95%CI	F	p
Blood pool
	SUVmean	0.93	[0.90–0.96]	5.06	0.03
	SULmean-James	0.92	[0.88 0.96]	5.15	0.03
	SULmean-Janma	0.94	[0.90–0.96]	4.74	0.04
Right liver lobe
	SUVmean	0.95	[0.93–0.97]	0.18	0.67
	SULmean-James	0.98	[0.97–0.99]	3.99	0.053
	SULmean-Janma	0.98	[0.97–0.99]	4.19	0.047
Left liver lobe
	SUVmean	0.86	[0.79–0.92]	0.22	0.64
	SULmean-James	0.86	[0.79–0.92]	0.34	0.56
	SULmean-Janma	0.89	[0.83–0.93]	0.44	0.51
Spleen
	SUVmean	0.98	[0.97–0.99]	4.44	0.04
	SULmean-James	0.92	[0.87–0.95]	2.88	0.10
	SULmean-Janma	0.93	[0.89–0.96]	2.77	0.10
Bone marrow
	SUVmean	0.92	[0.88–0.95]	3.06	0.09
	SULmean-James	0.95	[0.92–0.97]	0.31	0.58
	SULmean-Janma	0.96	[0.94–0.98]	0.32	0.57
Muscle
	SUVmean	0.78	[0.66–0.87]	5.04	0.03
	SULmean-James	0.75	[0.61–0.85]	5.20	0.03
	SULmean-Janma	0.78	[0.65–0.87]	4.97	0.03

ICC: Intraclass Correlation Coefficient; CI: confidence interval. SUV: standardized uptake value or SUV normalized by body weight. SUL, SUV normalized by lean body mass. F and p represent the test statistics and significance level of the Two-way Repeated-measures ANOVA

Figure 3; Table 2 illustrate the distribution of SUV/SUL values across different tissues. The degree of data dispersion varies according to the normalization method. The CQV values of SUVmean, SULmean-James, and SULmean-Janma for each candidate reference tissue were as follows: blood pool (CQV = 15.4%, 10.9%, 12.0%, respectively), right liver lobe (CQV = 12.7%, 10.6%, 12.3%, respectively), left liver lobe (CQV = 10.3%, 9.4%, 14.1%, respectively), spleen (CQV = 22.2%, 19.9%, 21.4%, respectively), bone marrow (CQV = 14.5%, 15.8%, 18.8%, respectively), and muscle (CQV = 15.7%, 13.2%, 15.5%, respectively). Overall, SUL-James exhibited the lowest CQV in most tissues, indicating the highest stability. When the Janma-based LBM correction was excluded, the spleen showed the greatest variability (CQV 19.9%-22.2%), reflecting substantial inter-individual differences. Variability across the remaining tissues was relatively comparable, with the blood pool (10.9%-15.4%) and liver (9.4%-12.7%) demonstrating the greatest stability.

Fig. 3.

Box plots of Body Weight (BW)- and Lean Body Mass (LBM)-corrected SUVmean/SULmean across background organs

Table 2.

Distribution of standardized uptake value measurements across different organs

Characteristic	Organs	Median (Q25, Q75)	CQV
SUVmean
	Blood pool	1.77 (1.57, 2.14)	15.4%
	Right liver lobe	1.20 (1.07, 1.38)	12.7%
	Left liver lobe	1.22 (1.09, 1.34)	10.3%
	Spleen	5.20 (4.09, 6.43)	22.2%
	Bone marrow	2.31 (2.00, 2.68)	14.5%
	Muscle	0.50 (0.43, 0.59)	15.7%
SULmean-James
	Blood pool	1.34 (1.19, 1.48)	10.9%
	Right liver lobe	0.91 (0.80, 0.99)	10.6%
	Left liver lobe	0.89 (0.82, 0.99)	9.4%
	Spleen	3.98 (3.09, 4.63)	19.9%
	Bone marrow	1.69 (1.49, 2.05)	15.8%
	Muscle	0.38 (0.33, 0.43)	13.2%
SULmean-Janma
	Blood pool	1.27 (1.10, 1.40)	12.0%
	Right liver lobe	0.88 (0.75, 0.96)	12.3%
	Left liver lobe	0.85 (0.73, 0.97)	14.1%
	Spleen	3.66 (2.86, 4.42)	21.4%
	Bone marrow	1.61 (1.38, 2.02)	18.8%
	Muscle	0.35 (0.30, 0.41)	15.5%

CQV: the coefficient of quartile variation, calculated as a percentage based on the interquartile range and median. SUV: standardized uptake value normalized by body weight. SUL, standardized uptake value normalized by lean body mass

Discussion

Derivatives of PET, such as SUVmean and SUVmax, are not only valuable for disease diagnosis but are also increasingly applied in treatment response assessment and prognostic prediction. In this study, we systematically evaluated different SUV normalization approaches and candidate reference tissues in the quantitative analysis of [⁶⁸Ga]Ga-Pentixafor PET/CT. The main findings are as follows: (1) SUL calculated using the James formula effectively reduced the body weight dependency observed with BW-normalized SUV, with the most pronounced effect in blood pool tissues; (2) among the candidate reference tissues, both liver and blood pool demonstrated low variability and high interobserver agreement, indicating their suitability as reliable reference tissues. These findings provide new evidence to support the establishment of standardized quantification and image interpretation frameworks for [⁶⁸Ga]Ga-Pentixafor PET/CT.

The search for optimal SUV normalization is not novel. Prior studies on [¹⁸F]F-FDG, [¹⁸F]F-MFBG, [⁶⁸Ga]Ga-PSMA, and [⁶⁸Ga]Ga-DOTATATE PET/CT have repeatedly compared BW- and LBM-normalized approaches. Evidence has shown that BW-normalized SUV is more susceptible to body fat composition when using tracers with low adipose uptake, whereas LBM-normalized SUL can significantly reduce this interference, particularly in longitudinal follow-up and multicenter settings [6–9, 18]. For [¹⁸F]F-NaF, which is almost exclusively distributed in the skeleton, SUV normalized to skeletal volume has proven to be more stable and representative than BW normalization [19]. Based on these experiences, we hypothesized that [⁶⁸Ga]Ga-Pentixafor PET/CT might similarly benefit from LBM correction. Our results confirmed this hypothesis: SUL derived from the James formula showed markedly reduced body weight dependency and lower variability in tissues such as blood pool and liver. In contrast, SUL derived from the Janma formula even demonstrated stronger body weight dependence in certain tissues, suggesting that the choice of LBM estimation method substantially impacts normalization outcomes. We speculate that the “idealized” LBM from the James formula may serve as a more appropriate scaling factor, better eliminating body size-related bias.

LBM estimation is not restricted to predictive equations; imaging-based methods have also gained attention. CT can estimate LBM by combining tissue volume with density, while MRI can evaluate water/fat fraction. Zhao et al. [7] demonstrated that Dixon-based MRI yielded the closest results to dual-energy X-ray absorptiometry, making it the most accurate imaging-based method available. However, Devriese et al. [20] reported that CT- and formula-based methods could differ significantly at the population level. Despite their accuracy, imaging-based LBM estimation has limited clinical applicability, requiring high-resolution scans and computational resources. Moreover, whole-body imaging is not routinely performed in all patients. Thus, formula-based LBM estimation remains the most practical approach in clinical settings.

With respect to reference tissue selection, the ideal candidate should demonstrate low variability, high reproducibility, and a relatively stable uptake level, thereby serving as a reliable baseline for lesion comparison. Our results indicate that both liver and blood pool met these criteria, showing high reproducibility across repeated measurements and uptake levels well above the reported range of lesion SUVs, supporting their suitability as reference tissues [21]. In the absence of suspected liver involvement, the liver should be prioritized, with the blood pool as a viable alternative. Notably, previous studies have suggested that [⁶⁸Ga]Ga-Pentixafor is not affected by significant tumor sink effects, further reinforcing the stability of these tissues as reference standards in longitudinal evaluations. In this study, we do not recommend using the spleen or bone marrow as routine reference organs. Previous research has shown that tracer uptake in both tissues can fluctuate during treatment in patients with multiple myeloma. For example, a decline in splenic uptake after first-line chemotherapy has been associated with poor prognosis, whereas baseline splenic activity prior to treatment shows no such relationship. Moreover, the uptake levels of these organs exhibit substantial heterogeneity—driven not only by interindividual differences but also by variations in disease burden and hematopoietic function. Patients with hematologic malignancies inherently display considerable variability in bone marrow activity, and even in the relatively healthy cohort of the present study, we observed between-subject variation in splenic and marrow uptake. Collectively, these findings indicate that neither organ provides the stability required for reliable normalization [12, 22].

We also explored the associations between sex, leukocyte counts, and SUV parameters. Interestingly, male patients consistently exhibited higher SUL values across multiple tissues compared to females. The underlying cause of this difference remains unclear but may relate to higher body weight and larger injected doses in males. Bone marrow uptake showed a weak correlation with leukocyte counts, suggesting that inflammatory status may influence [⁶⁸Ga]Ga-Pentixafor distribution. However, the clinical significance of these findings requires further validation in larger cohorts.

This study has several limitations. First, the sample size was relatively small, and only patients with primary aldosteronism were included; thus, generalization to oncologic populations requires further confirmation. Second, our analysis was restricted to static imaging at a single time point (40–60 min), without dynamic imaging data to capture temporal uptake changes.

Conclusion

In conclusion, our findings recommend the use of SUL calculated with the James formula, together with the liver or blood pool as reference tissues, to improve the stability and comparability of quantitative analysis in [⁶⁸Ga]Ga-Pentixafor PET/CT. Future studies with larger sample sizes and multicenter settings are warranted to further validate these results and to explore their applications in treatment response assessment, prognostic prediction, and inter-center harmonization.

Author contributions

All authors contributed to the study conception and design. Material preparation, data collection and analysis were performed by YGL; RBT, HYY and SJT. The first draft of the manuscript was written by YGL and RBT. YC provided some valuable suggestions. The draft was revised by LQ and ZWH. All authors read and approved of the final manuscript.

Funding

This study was partially supported by the scientific and technological project of the Health Commission of Sichuan Province (Project No. 24QNMP093) and Special Funding for Cultivation of Defense Talents (CW202410).

Data availability

The datasets generated during and/or analyzed during the current study are available from the corresponding author upon reasonable request.

Declarations

Ethics approval

This study was conducted in accordance with the Declaration of Helsinki and approved by the Institutional Review Board (IRB) of Southwest Medical University. Clinical trial number: Not applicable.

Consent to participate

Written informed consents were obtained from the included subjects for participation in this study.

Consent for publication

The authors affirm that all human participants involved in this study have signed the consent form and agreed to the publication of their medical imaging.

Competing interests

The authors declare no competing interests.