New PET Scan Model Could Tell Dangerous Childhood Tumors From Benign Ones Without Surgery

For parents of children diagnosed with a neuroblastic tumor, one of the most agonizing uncertainties begins the moment the tumor is found: is it an aggressive neuroblastoma that demands intensive chemotherapy, or a more indolent ganglioneuroblastoma that may need far less intervention? The answer currently requires a biopsy, an invasive procedure that carries real risks in small children whose tumors often sit deep in the abdomen or chest, wrapped around critical blood vessels and nerves. Now a team of nuclear medicine physicians in Beijing has shown that a specialized PET scan, combined with a handful of routine clinical facts, may be able to make that distinction noninvasively, offering a glimpse of a future where some children could be spared the scalpel altogether.

The study, published in the European Journal of Nuclear Medicine and Molecular Imaging by Xiaoya Wang, Keyu Zhang and colleagues at Beijing Friendship Hospital, Capital Medical University, focused on a radiotracer called ¹⁸F-meta-fluorobenzylguanidine, or ¹⁸F-MFBG. The tracer is a fluorine-18-labeled analogue of the long-standing imaging agent MIBG, and it exploits a quirk of neuroblastoma biology: these tumors arise from the sympathetic nervous system and, like their healthy counterparts, express the norepinephrine transporter on their cell surfaces. When injected into a patient, ¹⁸F-MFBG travels through the bloodstream and is actively pumped into tumor cells by this transporter, lighting up the tumor on a PET scan with a specificity that few other tracers can match.

The researchers enrolled 42 pediatric patients, all with histologically confirmed diagnoses established after the scan: 23 with neuroblastoma and 19 with ganglioneuroblastoma, the intermediate form of the disease that contains a mixture of primitive neuroblasts and more mature ganglion cells. Each child underwent ¹⁸F-MFBG PET/CT before any treatment began, allowing the team to extract quantitative imaging parameters from the primary tumors without the confounding effects of chemotherapy or surgery. Alongside the imaging data, the team compiled clinical variables including age at diagnosis, the presence of bone metastases, and genetic findings such as the 11q23 deletion, a chromosomal abnormality associated with aggressive disease.

The quantitative analysis went well beyond simple visual inspection. The team measured standard uptake values, including SUVmax and SUVmean, which reflect how avidly the tumor traps the tracer, as well as tumor-to-background ratios that normalize uptake against surrounding tissue. They also computed texture-based features such as AUC-CSH, a measure derived from the histogram of voxel intensities that captures the heterogeneity of tracer distribution within the tumor, and TL-hNET, an estimate of total lesion norepinephrine transporter expression that integrates uptake intensity with metabolic tumor volume. Six of these parameters, including SUVmax, SUVmean, TBRmean, TBRmax, AUC-CSH and TL-hNET, differed significantly between neuroblastoma and ganglioneuroblastoma, confirming that the two tumor types handle the tracer in measurably different ways.

Those differences make biological sense. Neuroblastoma, composed of densely packed, primitive neuroblasts, tends to express the norepinephrine transporter at high levels and take up MFBG avidly. Ganglioneuroblastoma, with its greater proportion of mature ganglion cells and Schwannian stroma, typically shows weaker transporter activity and therefore dimmer signal on the scan. But the researchers also encountered a familiar statistical obstacle: the imaging parameters were heavily intercorrelated, with variance inflation factors exceeding 10, meaning they carried largely redundant information. Rather than stuffing all of them into a model and risking overfitting, the team selected SUVmax as the single representative imaging variable for multivariable analysis, a disciplined choice given the modest sample size.

To handle the small cohort without producing unstable estimates, the researchers turned to Firth logistic regression, a penalized form of logistic regression specifically designed for situations where events are rare or predictors are numerous relative to the sample. In the final multivariable model, two variables emerged as independent predictors of neuroblastoma: SUVmax, with an odds ratio of 8.665, meaning each increment in tracer uptake sharply raised the odds of malignancy, and younger age, with an odds ratio of 0.676, reflecting the well-established fact that neuroblastoma tends to strike earlier in life than its better-differentiated counterpart. Bone metastasis showed a strong trend toward independent association, with an odds ratio of 11.753 and a p-value of 0.050, hovering just at the threshold of conventional statistical significance.

With these predictors in hand, the team constructed and compared three diagnostic models: an imaging-only model using SUVmax alone, a clinical model combining age and bone metastasis, and an integrated model merging all three variables. The results told a clear story of incremental value. The imaging-only model achieved an area under the ROC curve of 0.810, the clinical model 0.787, and the integrated model 0.856, with a 95 percent confidence interval of 0.737 to 0.975. While the differences in AUC alone showed only a trend toward significance on the DeLong test, two complementary statistical tools provided firmer support. Bootstrap validation across 2,000 resamples confirmed the stability of the estimates, and integrated discrimination improvement analysis showed that the combined model significantly outperformed both alternatives, with an IDI of 0.122 against the imaging-only model and 0.083 against the clinical model, both statistically significant.

The clinical implications of these numbers are worth unpacking. An AUC of 0.856 means the integrated model discriminates between the two tumor types correctly in roughly 86 percent of paired comparisons, a level of performance that, if confirmed in larger cohorts, could meaningfully change the diagnostic pathway for children with neuroblastic tumors. A child whose scan shows high SUVmax, who is very young, and who has bone metastases would be flagged as high-risk for neuroblastoma, potentially accelerating the decision to begin treatment. Conversely, a child with low tracer uptake, older age and no skeletal spread could be triaged toward a more watchful approach, sparing families the physical and emotional toll of an unnecessary biopsy in a location where complications could be severe.

The study’s authors are appropriately measured in their conclusions, framing the work as a preliminary hypothesis-generating effort that awaits rigorous external validation in larger, multicenter cohorts. That caution is warranted. Forty-two patients is a small sample, and the model was developed and evaluated within the same dataset, a design that inflates apparent performance compared with true out-of-sample prediction. The field of radiomics and imaging-based diagnostics is littered with promising single-center studies that failed to replicate, a problem that has prompted formal criteria for translating such models into clinically useful tests. The Beijing team’s transparent reporting of confidence intervals, bootstrap validation and reclassification statistics suggests an awareness of these pitfalls, but the ultimate test will be independent replication.

Still, the direction of travel is unmistakable. ¹⁸F-MFBG offers practical advantages over the iodine-123-labeled MIBG that has dominated neuroblastoma imaging for decades, including superior spatial resolution, the ability to perform same-day imaging, and no requirement for thyroid blockade or iodine saturation protocols. Combined with quantitative texture analysis and thoughtful integration of clinical and genetic risk factors, it is becoming a probe not just of where a tumor is, but of what it fundamentally is. For a disease whose treatment intensity spans the full spectrum from observation to myeloablative therapy, that kind of molecular discrimination, delivered through a simple scan rather than a surgical needle, could reshape pediatric cancer care. The next step, larger multicenter trials, will determine whether this preliminary signal hardens into a standard of practice.

Subject of Research: Noninvasive differentiation of neuroblastoma from ganglioneuroblastoma using ¹⁸F-MFBG PET/CT imaging parameters and clinical factors

Article Title: Diagnostic performance of multiparametric model integrating ¹⁸F-MFBG PET/CT imaging parameters and clinical factors for differentiating neuroblastoma from ganglioneuroblastoma: a preliminary study

Article References: Wang, X., Zhang, K., Kan, Y., Wang, W., & Yang, J. (2026). Diagnostic performance of multiparametric model integrating ¹⁸F-MFBG PET/CT imaging parameters and clinical factors for differentiating neuroblastoma from ganglioneuroblastoma: a preliminary study. European Journal of Nuclear Medicine and Molecular Imaging. https://doi.org/10.1007/s00259-026-08185-7

Image Credits: AI Generated

DOI: 10.1007/s00259-026-08185-7

Keywords: neuroblastoma, ganglioneuroblastoma, 18F-MFBG, PET/CT, pediatric oncology, molecular imaging, SUVmax, diagnostic model, norepinephrine transporter, logistic regression, radiomics, biopsy