Translate this page into:
Optimising FDG-PET/CT Segmentation in Metastatic Breast Cancer: Comparison of SUVmax Thresholds Using syngo.via
, Anjali Jain1, Sharjeel Usmani1, Khulood Al Riyami1, Rashid Alsukaiti1
*Corresponding author: Dr. Subhash Chand Kheruka, Department of Radiology and Nuclear Medicine, Sultan Qaboos Comprehensive Cancer Care and Research Centre, University Medical City, Muscat, Oman. skheruka@gmail.com
-
Received: ,
Accepted: ,
How to cite this article: Kheruka SC, Al-Rashdi S, Al-Maymani N, Raii T, Al-Makhmari N, Al-Saidi H, et al. Optimising FDG-PET/CT Segmentation in Metastatic Breast Cancer: Comparison of SUVmax Thresholds Using syngo.via. Indian J Nucl Med. 2026;41:320-6. doi: 10.25259/IJNM_35_2026
Abstract
Objectives:
Semi-automated lesion delineation in fluorodeoxyglucose-positron emission tomography/computed tomography (FDG-PET/CT) is sensitive to the maximum standardised uptake value (SUVmax) threshold used, but its effect on quantitative outputs and image-quality surrogates is not standardised. We compared commonly used thresholds to identify a practical setting for workstation-based SUVmax-threshold segmentation in syngo.via, within a PERCIST-consistent clinical workflow.
Material and Methods:
We retrospectively reviewed FDG-PET/CT scans from 37 women with metastatic breast cancer, each with a single metastatic lesion. Lesions were contoured in syngo.via using fixed isocontours at 20%, 30%, 40% and 50% of lesion SUVmax. For each threshold, we calculated SUVmean, metabolic tumour volume (MTV), total lesion glycolysis (TLG), signal-to-noise ratio (SNR), and contrast-to-noise ratio (CNR). Medians and percentage changes were summarised, and trends were visualised with line charts. Differences across thresholds were assessed using Friedman tests; associations between metrics were examined with Spearman correlation.
Results:
Increasing the threshold from 20% to 50% increased median SUVmean by 52% (3.35 to 5.09), SNR by 59% (52.00 to 82.64), and CNR by 62% (48.25 to 78.09), while reducing MTV by 81% (4.51 to 0.84 cm3) and TLG by 69% (14.35 to 4.49) (all p<0.001). SUVmean, SNR, and CNR were strongly positively correlated (ρ>0.9), whereas MTV and TLG were strongly negatively correlated (ρ<−0.9) as the threshold increased. The 40% threshold provided a workable middle ground, improving conspicuity without an excessive loss of volumetric burden.
Conclusion:
A 40% SUVmax threshold appears to offer the most balanced overall workstation-based setting in syngo.via, supporting lesion detection while preserving clinically meaningful estimates of disease extent for staging and treatment monitoring.
Keywords
Contrast-to-noise ratio
FDG-PET/CT
Lesion segmentation
Metabolic tumour volume
Signal-to-noise ratio
SUVmax threshold
INTRODUCTION
18F-fluorodeoxyglucose positron emission tomography/computed tomography (FDG PET/CT) has become an indispensable tool in oncology, providing unparalleled insight into tumour biology by measuring metabolic activity rather than relying solely on anatomical size. In addition to its established role in staging and treatment planning, FDG PET/CT is increasingly used for therapy response evaluation, complementing traditional morphologic criteria such as RECIST and advancing functional assessment through the PERCIST framework.[1,2] PERCIST is primarily a response assessment framework based on lean-body-mass–corrected uptake (SULpeak) and standardised reference measurements; it does not prescribe SUVmax-percentage thresholding for lesion delineation. In routine clinical practice, however, workstation-based semi-automated segmentation often relies on fixed SUVmax isocontours to derive volumetric and intensity-based metrics. These PET-based response criteria allow clinicians to assess metabolic changes more sensitively than anatomical imaging alone, enabling earlier and more reliable decisions in patient management.
Quantitative PET parameters such as standardised uptake value, metabolic tumour volume, and total lesion glycolysis are important markers of disease activity and treatment response. Image quality indicators, including signal-to-noise ratio and contrast-to-noise ratio, also play a critical role in ensuring accurate lesion detection and consistent quantification.[3,4] A key challenge in PET analysis is defining the appropriate volume of interest. Thresholding based on SUVmax is widely applied because of its simplicity and reproducibility; however, the optimal threshold percentage remains uncertain. Lower thresholds may include background noise and physiologic uptake, whereas higher thresholds may exclude viable tumour tissue, with both scenarios potentially affecting clinical interpretation. These effects can be amplified in smaller lesions because of partial volume effects and, depending on lesion location, background activity and motion (for example, in lung or liver), may further influence measured MTV and TLG.
Several studies have investigated the influence of SUVmax-based threshold selection on PET-derived metrics and lesion segmentation. Boellaard et al.[3] emphasised the importance of standardised imaging protocols to reduce variability between institutions. Vanderhoek et al.[5] demonstrated that small changes in threshold selection could substantially alter metabolic tumour volume and total lesion glycolysis measurements, thereby influencing treatment response interpretation. Schaefer et al.[6] proposed contrast-oriented segmentation techniques to improve delineation accuracy, while Lodge[7] and Kinahan[8] highlighted standardisation as a cornerstone of reliable PET quantification. Despite these efforts, consensus on the optimal SUVmax threshold has not been established, particularly in advanced disease where lesion size, location, and background activity vary considerably.
In metastatic breast cancer, common metastatic sites include bone, lymph nodes, liver, and lungs, regions in which accurate lesion detection and quantification are essential for therapeutic decision making.[9] For longitudinal PET/CT assessment, applying a consistent segmentation approach is important to minimise observer- and method-related variability, while recognising that PERCIST itself is anchored to SUL peak rather than SUVmax isocontour delineation. This study systematically evaluates four fixed SUVmax thresholds, namely 20%, 30%, 40% and 50%, in FDG PET/CT scans from 37 women with metastatic breast cancer, each presenting a single measurable lesion. The effects of threshold selection on SUVmean, metabolic tumour volume, total lesion glycolysis, signal-to-noise ratio, and contrast-to-noise ratio were assessed using syngo.via software (Siemens Healthineers). The objective of this work is to identify the SUVmax threshold that provides the most practical balance between lesion visibility and disease quantification, offering guidance for workstation-based standardised PET analysis in clinical practice.
MATERIAL AND METHODS
Study population
This retrospective analysis included a cohort of 37 female patients (mean age ± SD: 55 ± 12 years) with histologically confirmed metastatic breast cancer, each presenting a single measurable metastatic lesion. Lesions were distributed across typical metastatic sites, with 40% in bone, 30% in lymph nodes, 20% in the liver, and 10% in the lungs. For each case, lesion size was recorded on the CT component as the maximum axial diameter (mm) (short-axis diameter for lymph nodes, where applicable).
All scans were anonymised and analysed under institutional review board approval, with a waiver of written consent. Eligibility required an FDG-avid lesion that was measurable on CT (≥10 mm) or clearly discernible on FDG PET, to support reliable quantitative analysis and reduce partial volume–related bias. Exclusion criteria included incomplete PET datasets, uncontrolled glucose levels, or scans acquired using non-standard imaging protocols.
Imaging protocol
All patients underwent whole-body 18F-FDG PET/CT imaging on a Siemens Biograph Vision 600 system (Siemens Healthineers, Erlangen, Germany). Patients fasted for at least six hours before tracer injection, and serum glucose was confirmed to be <150 mg/dL. A standard activity of 5.5 MBq/kg (±10%) of 18F-FDG was administered intravenously, followed by a 60-minute uptake period in a quiet, dimly lit environment.
PET data were acquired in three-dimensional mode for two to three minutes per bed position and reconstructed using iterative OSEM with time-of-flight (Iterative+TOF; four iterations, five subsets), with attenuation correction based on low-dose CT. Images were reconstructed with an image matrix of 220×220 (image size 220; zoom 1.0) and a 4-mm FWHM Gaussian post-reconstruction filter. CT acquisition parameters were 120 kVp with CARE Dose4D, 128×0.6 mm collimation, and 5-mm reconstructed slice thickness. Images were reviewed on syngo.via VB30 software in axial, coronal, and sagittal planes.
VOI segmentation and quantitative metrics
Volumes of interest (VOIs) were automatically segmented using syngo.via software, based on fixed SUVmax-derived thresholds of 20%, 30%, 40%, and 50%. For each threshold level, voxels with SUV values equal to or exceeding the specified percentage of the lesion SUVmax were included in the VOI. This standardised method allowed consistent comparison of how threshold selection influenced both quantitative measurements and image quality parameters.
The following metrics were extracted for each VOI:
SUVmean: Mean standardised uptake value within the VOI, representing the average metabolic activity of the segmented lesion.
MTV (cm3): Metabolic tumour volume, defined as the volume of metabolically active tumour tissue.
TLG: Total lesion glycolysis, calculated as SUVmean multiplied by MTV, integrating both metabolic intensity and lesion volume.
SNR (signal-to-noise ratio): Calculated as the lesion SUVmean divided by the standard deviation (SD) of the background, indicating lesion signal strength relative to image noise.
CNR (contrast-to-noise ratio): Defined as (SUVmean of lesion minus SUVmean of background) divided by SD of background, representing lesion conspicuity against surrounding tissue uptake.
Background SUVmean and SD were measured within a homogeneous reference region, preferentially the liver parenchyma, using a 3-cm-diameter spherical VOI, in accordance with EANM and PERCIST guidelines.[1,3] The liver was selected as the primary reference region due to its relatively uniform and physiologically stable FDG uptake, low interpatient variability, and established validity in PET quantification. Using the liver as a reference improves normalisation of lesion uptake, reduces background variability, and enhances the reproducibility of semi-quantitative parameters such as SNR, CNR, MTV, and TLG, particularly in longitudinal evaluations.
In cases where hepatic involvement or abnormal liver uptake was observed, an alternative homogeneous region, such as the mediastinum, was used. All measurements were performed by two experienced observers, with discrepancies resolved through consensus to ensure consistency and minimise interobserver variability.
RESULTS
Data from 37 metastatic lesions are summarised in Table 1 (median values).
| Metric | 20% Threshold | 30% Threshold | 40% Threshold | 50% Threshold |
|---|---|---|---|---|
| SUVmean | 3.35 (2.80–3.95) | 4.05 (3.40–4.70) | 4.69 (3.95–5.30) | 5.09 (4.30–5.80) |
| MTV (cm3) | 4.51 (2.10–8.90) | 2.15 (1.20–4.30) | 1.58 (0.90–3.10) | 0.84 (0.40–1.80) |
| TLG | 14.35 (7.20–28.60) | 9.15 (4.80–17.50) | 7.22 (3.90–13.80) | 4.49 (2.10–9.20) |
| SNR | 52.00 (40.10–65.30) | 63.17 (50.20–75.80) | 70.50 (58.40–85.60) | 82.64 (65.10–98.30) |
| CNR | 48.25 (36.80–60.40) | 58.17 (45.30–70.20) | 70.00 (55.60–82.90) | 78.09 (62.40–95.10) |
Values are presented as median (IQR). Thresholds are expressed as % of lesion SUVmax. SNR = lesion SUVmean/background SD; CNR = (lesion SUVmean − background SUVmean)/background SD. MTV: Metabolic tumour volume; TLG: Total lesion glycolysis; FDG-PET: Fluorodeoxyglucose-positron emission tomography; SUVmax: Maximum standardised uptake value; SNR: Signal-to-noise ratio, CNR; Contrast-to-noise ratio , SUVmean: Median standardised uptake value, MTV: Metabolic tumour volume
As the maximum standardised uptake value (SUVmax) threshold increased from 20% to 50%, the corresponding median standardised uptake value (SUVmean) rose by approximately 52% (from 3.35 to 5.09), the signal-to-noise ratio (SNR) increased by 59% (from 52.00 to 82.64), and the contrast-to-noise ratio (CNR) increased by 62% (from 48.25 to 78.09). In contrast, the metabolic tumour volume (MTV) decreased by 81% (from 4.51 to 0.84 cm3), and the total lesion glycolysis (TLG) decreased by 69% (from 14.35 to 4.49).
These findings indicate that as the SUVmax threshold becomes more stringent, the segmented tumour region becomes smaller but more focused on areas of higher metabolic activity, yielding a cleaner and more specific tumour signal. Overall, the results demonstrate that threshold selection for lesion segmentation can significantly influence the quantitative metrics derived from FDG PET imaging. This underscores the importance of careful threshold calibration to ensure accurate metabolic assessment in oncologic evaluations.
Statistical analysis
The Friedman test revealed statistically significant differences in results across all SUVmax thresholds (p < 0.001) [Friedman χ2(3) = 87.1–88.8; Kendall’s W = 0.91–0.92]. Spearman correlation analysis demonstrated strong positive correlations between threshold level and SUVmean (rho = 0.95, 95% CI 0.88–1.00), SNR (rho = 0.92, 95% CI 0.87– 1.00), and CNR (rho = 0.94, 95% CI 0.87–1.00). In contrast, MTV (rho = −0.96, 95% CI −1.00 to −0.88) and TLG (rho = −0.93, 95% CI −1.00 to −0.88) showed strong negative correlations (all p < 0.001). These findings confirm that as the SUVmax threshold increases, signal-based parameters tend to increase, while volumetric measurements decrease.
Scatter plots of individual lesion data revealed some interpatient variability; however, the overall trends remained consistent. A few outliers exhibited notably high MTV values at lower thresholds, likely due to inclusion of background or low-uptake regions in the VOI.
Visualisation of threshold effects
Fig 1 to 5 collectively illustrate how each quantitative metric—SUVmean, MTV, TLG, CNR, and SNR—varies with increasing SUVmax thresholds in patients with metastatic breast cancer presenting a single metastatic lesion. As shown in Fig 1, SUVmean demonstrates a consistent upward trend, indicating that higher thresholds isolate the most metabolically active regions of the lesion. These FDG-avid areas dominate the signal once background tissue and lowuptake voxels are excluded from the segmentation.
Conversely, Fig 2 and 3 show progressive reductions in MTV and TLG as the threshold increases. This inverse relationship highlights how higher thresholds compress the volume of interest to the metabolic core of the lesion, reducing background inclusion and improving the specificity of metabolic quantification. Such refinement is particularly important in anatomically complex regions commonly involved in breast cancer metastases, including the liver, lymph nodes, and skeletal system.
Fig 4 and 5 reveal corresponding improvements in contrast-to-noise ratio (CNR) and signal-to-noise ratio (SNR). These enhancements suggest that lesion conspicuity and overall image clarity improve with higher thresholds, allowing for more confident identification— especially in small or borderline lesions that might otherwise be masked by physiological uptake or background signal.
Taken together, these findings illustrate a distinct trade-off. Lower thresholds provide a broader volumetric representation of disease burden, but at the cost of reduced specificity due to inclusion of background activity. In contrast, higher thresholds enhance image quality and signal clarity but may underestimate the full extent of tumour involvement. Among the thresholds evaluated, the 40% SUVmax cutoff appears to strike the most practical balance—retaining sufficient lesion volume for reliable quantification while delivering improved image clarity and reproducibility. In this context, “balance” refers to improved CNR/SNR without a disproportionate reduction in MTV/TLG at higher thresholds. This makes it particularly well-suited for a PERCIST-consistent clinical workflow and serial monitoring, noting that PERCIST itself is based on SULpeak rather than SUVmax-percentage lesion delineation.
A representative example of VOI placement and threshold-based segmentation is shown in Fig 6. The metastatic lesion VOI was generated using syngo.via software with fixed SUVmax percentage thresholds of 20%, 30%, 40%, and 50%, enabling direct visual comparison of how higher thresholds progressively limit the VOI to the most intensely FDG-avid tumour core while excluding lower-uptake peripheral regions. For background normalisation, a liver reference VOI was positioned in the right hepatic lobe within a visually homogeneous area of parenchyma, avoiding vessels and focal abnormalities, in line with EANM guidelines and PERCIST reference region recommendations. This illustrative case complements the quantitative analysis by demonstrating how threshold selection affects lesion delineation and the derived metabolic measurements in practical terms.
DISCUSSION
This study demonstrates that selecting the SUVmax threshold significantly influences quantitative FDG PET metrics in metastatic breast cancer, affecting lesion detectability and treatment response evaluation in a PERCIST-consistent clinical workflow (noting that PERCIST response assessment is based on SULpeak rather than SUVmax-percentage lesion delineation). Increasing the threshold from 20% to 50% led to consistent increases in SUVmean, SNR, and CNR, reflecting improved identification of metabolically active tumour regions. Simultaneously, MTV and TLG decreased, indicating tighter and more specific delineation of tumour boundaries. These results, supported by strong correlations and statistical significance (p < 0.001), highlight the critical importance of threshold optimisation for consistent PET quantification and interpretation.
Threshold adequacy for workstation-based segmentation in a PERCIST-consistent workflow
This analysis identifies a 40% SUVmax threshold as the most reliable balance between lesion visibility and accurate disease burden estimation. At this threshold, CNR and SNR increased significantly compared to lower levels, ensuring clear visualisation of lesions in lymph nodes, bone, and lungs. While a 50% threshold provided the highest signal quality, it frequently excluded lower uptake regions, potentially underestimating lesion volume and metabolic activity. Conversely, thresholds below 30% were more likely to include physiologic uptake and background noise, reducing specificity.
Therefore, the 40% cutoff produces consistent and reproducible VOIs suitable for both baseline and follow-up evaluations when applying a consistent segmentation approach for serial monitoring. Importantly, the goal of applying a consistent threshold is not to fix SUVmax itself, which may vary physiologically or with treatment, but rather to maintain methodological consistency. This ensures that relative changes in SUVmean, MTV, or TLG reflect actual biological response rather than differences in segmentation technique. These findings align with prior research advocating for standardisation in PET quantification [3,4]and build upon earlier work by Pinker et al.[10] which demonstrated the prognostic significance of TLG in breast cancer. In this study, “balance” refers to improved CNR/SNR without a disproportionate reduction in MTV/TLG at higher thresholds.
Clinical implications
Clinically, applying an optimised threshold improves the reliability of staging and treatment response assessment. Enhanced CNR and SNR at the 40% threshold allow for the detection of smaller or subtle metastatic foci that may be overlooked at lower thresholds, leading to more accurate TNM classification. At the same time, consistent volumetric measurements such as MTV and TLG provide quantitative biomarkers that can inform decisions regarding treatment adjustments and prognosis.
Automated segmentation based on fixed thresholds using syngo.via reduces interobserver variability and supports reproducible assessments over time. Maintaining a consistent threshold across serial scans enhances confidence in detecting metabolic changes, particularly in early treatment phases when anatomical changes may still be minimal.
LIMITATIONS
The primary limitation of this study is the relatively small sample size (n = 37), which may limit generalisability. Nonetheless, the uniform trends observed across all cases support the robustness of the findings. The retrospective, single-centre design also restricted comparisons with other segmentation strategies, such as adaptive or gradient-based methods.
Future directions
Moreover, site-specific variations—such as bone versus liver metastases—may affect the optimal threshold, but were not analysed separately in this study. Future research should confirm these results in larger, multicentre cohorts and explore the relationship between threshold-dependent PET metrics and clinical outcomes, including progression-free and overall survival. Additionally, investigating the utility of adaptive or machine learning-based thresholding techniques may further enhance segmentation accuracy and reliability across diverse lesion types and imaging systems. Finally, this study did not include a ground-truth reference (e.g., expert manual contours or phantom-validated thresholds). Therefore, the findings primarily describe how threshold choice influences derived metrics rather than absolute delineation accuracy.
CONCLUSION
This study emphasises the clinical significance of selecting an appropriate SUVmax threshold in FDG PET/CT for patients with metastatic breast cancer. Consistent with prior evidence that standardised quantification improves reproducibility and lesion detectability, higher thresholds enhanced visual clarity by increasing SUVmean, SNR, and CNR, while lower thresholds offered a more comprehensive representation of total disease burden through increased MTV and TLG. Among the thresholds tested, 40% SUVmax emerged as the most clinically useful and balanced cutoff, optimising both lesion detection sensitivity and quantitative accuracy when using a consistent workstation-based segmentation approach for serial monitoring (recognising that PERCIST response assessment is based on SUL peak). These results support broader adoption of harmonised PET quantification methods to improve staging accuracy, inform individualised treatment strategies, and enable earlier detection of metabolic response to therapy.
Author contributions:
SCK: Conceptualisation, methodology, data curation, formal analysis, visualisation, writing—original draft; SA, NaA, TR, NoA, HA and AA: Data curation, image/VOI measurements, review and editing; AJ, SU, KA and RA: Clinical interpretation, supervision, review and editing. All authors approved the final manuscript and accepted accountability for the work.
Ethical approval:
Institutional Review Board approval is not required as it is a retrospective study.
Declaration of patient consent:
Patient’s consent not required as the patient's identity is not disclosed or compromised. All scans were anonymised and analysed under institutional review board approval, with a waiver of written consent.
Conflicts of interest:
There are no conflicts of interest.
Use of artificial intelligence (AI)-assisted technology for manuscript preparation:
The authors confirm that there was no use of artificial intelligence (AI)-assisted technology for assisting in the writing or editing of the manuscript and no images were manipulated using AI.
Financial support and sponsorship: Nil.
References
- From RECIST to PERCIST: Evolving considerations for PET response criteria in solid tumors. J Nucl Med. 2009;50:122S-150S.
- [CrossRef] [PubMed] [Google Scholar]
- New response evaluation criteria in solid tumours: Revised RECIST guideline (version 1.1) Eur J Cancer. 2009;45:228-47.
- [CrossRef] [PubMed] [Google Scholar]
- FDG PET/CT: EANM procedure guidelines for tumour imaging: Version 2.0. Eur J Nucl Med Mol Imaging. 2015;42:328-54.
- [CrossRef] [PubMed] [Google Scholar]
- Repeatability of SUV in oncologic 18F-FDG PET. J Nucl Med. 2017;58:523-32.
- [CrossRef] [PubMed] [Google Scholar]
- Impact of the definition of peak standardized uptake value on quantification of treatment response. J Nucl Med. 2012;53:37-43.
- [CrossRef] [PubMed] [Google Scholar]
- A contrast-oriented algorithm for FDG-PET-based delineation of tumour volumes for radiotherapy planning: Derivation from phantom measurements and validation in patient data. Eur J Nucl Med Mol Imaging. 2008;35:1989-99.
- [CrossRef] [PubMed] [Google Scholar]
- Practical considerations for PET response criteria in solid tumors (PERCIST) and beyond. PET Clin. 2016;11:211-22.
- [Google Scholar]
- Positron emission tomography-computed tomography standardized uptake values in clinical practice and assessing response to therapy. Semin Ultrasound CT MR. 2010;31:496-505.
- [CrossRef] [PubMed] [Google Scholar]
- PET/CT for patients with breast cancer: Where and when? Eur J Nucl Med Mol Imaging. 2019;46:1923-36.
- [Google Scholar]
- The impact that number of analyzed metastatic breast cancer lesions has on response assessment by 18F-FDG PET/CT using PERCIST. J Nucl Med. 2016;57:1102-1104.
- [CrossRef] [PubMed] [Google Scholar]