An International Study of Factors Affecting Variability of Dosimetry Calculations, Part 6: Lessons Learned, Achievable Variability, and the Road Ahead
In brief
Excluding inconsistent dosimetry submissions narrowed kidney dose variability from 23% to 3%
In a retrospective reanalysis of an international dosimetry challenge, excluding about half the full-workflow submissions for errors or inconsistent methods narrowed the total-kidney dose variability for one patient from 23.1% to 3.0%. Segmentation and time-activity curve methods drove much of the variation; standardization and quality checks could improve reproducibility, though some lesion estimates remained variable.
- Journal
- Journal of nuclear medicine : official publication, Society of Nuclear Medicine (Q1)
- Published
- 19 November 2026
- Study design
- Cohort / observational study
- Evidence level
- Level 3, Low (CEBM 3b)
- Authors
- Carlos Uribe, Sara Kurkowska, Julia Brosch-Lenz, Yuni Dewaraja, Eric Frey, John Sunderland
- PMID
- 43057880
- DOI
- 10.2967/jnumed.126.272798
Why clinicians should know about it
- Picked for Medical Physics (paper of the day, 22 November 2026): International Lu-177 challenge quantifies dosimetry variability and QA gains
Abstract
Personalized radiopharmaceutical therapies rely on accurate and reproducible absorbed dose estimates. However, substantial variability persists because of differences in imaging protocols, segmentation, time-activity curve fitting and integration, and dose calculation implementation. The Society of Nuclear Medicine and Molecular Imaging 177Lu Dosimetry Challenge was a 5-task international study designed to isolate sources of variability in image-based dosimetry workflows. In this final publication, we integrate findings across the series and perform a retrospective analysis of the original full-workflow submissions to estimate the reduction in variability achievable through methodologic consistency and quality assurance. Methods: In this final analysis, the original task 1 end-to-end SPECT/CT workflow submissions were reevaluated to generate a refined task 1 dataset by excluding previously identified processing or reporting mistakes, as well as submissions with methodologic choices inconsistent with the recommendations derived from the challenge. Exclusion criteria included nonstandard volume-of-interest (VOI) definitions, inconsistent VOI handling, inappropriate time-activity curve fitting and integration, and omission of required mass scaling. Variability was quantified using the interquartile range and quartile coefficient of dispersion (QCD). Results: The refined task 1 analysis showed that a substantial fraction of the observed variability is potentially avoidable. Exclusions reduced the analyzed dataset by approximately 50% and narrowed absorbed dose distributions for most organs and lesions. For patient A, QCD decreased for total kidney (23.1% to 3.0%), spleen (24.2% to 7.8%), and lesion 1 (22.8% to 13.3%). For patient B, organ variability also decreased (12.6% to 8.2% in the liver and 15.3% to 10.9% in the left kidney), although variability for some lesions decreased less substantially. Findings across the challenge indicated that, among the workflow steps evaluated, segmentation and time-activity curve fitting/integration were the dominant contributors to variability, whereas variability from the absorbed dose calculation step alone was generally small once upstream inputs were standardized. Conclusion: A substantial portion of variability in radiopharmaceutical therapy dosimetry is potentially avoidable through standardization and quality assurance. Under more standardized conditions, organ absorbed dose estimates showed markedly lower variability, with QCD values as low as 3%. These results support implementation of standardized VOI segmentation rules, time-activity curve modeling conventions, and auditable quality control and reporting procedures to improve reproducibility in both clinical trials and routine practice.
Abstract as published, via PubMed.
For healthcare professionals. The summary is generated by AI from the published abstract, and the evidence level is assigned automatically from the study design on the Oxford CEBM hierarchy. Neither is medical advice. Read the full paper before changing practice.