An international study of factors affecting variability of dosimetry calculations, part 5: impact of segmentation...

Individualized radiopharmaceutical therapies guided by patient-specific absorbed dose assessments using imaging have the potential to improve both efficacy and safety. Understanding sources of variability in absorbed dose calculations is critical for standardization. The Society of Nuclear Medicine and Molecular Imaging Dosimetry Task Force launched the ¹⁷⁷Lu Dosimetry Challenge to evaluate variability across different steps within the dosimetry workflow. This work aimed to assess the variability in absorbed doses due to differences in segmentation methods.

Anonymized datasets from two patients treated with ¹⁷⁷Lu-DOTATATE, including serial SPECT/CT scans, were made available online. Participants were asked to segment healthy organs and lesions and perform dosimetry calculations. In a subsequent task, participants were provided with standardized segmented VOIs and asked to perform dosimetry based on these pre-defined regions. Variability in segmentation was assessed by comparing absorbed dose estimates across two scenarios: participant-generated segmentations versus predefined reference segmentations. Relative absorbed dose variability was quantified using the quartile coefficient of dispersion (QCD) and interquartile range.

Variability in absorbed dose (measured as QCD difference between absorbed doses from participant-generated segmentations and those from reference segmentations) for kidneys was less than 5% in simple cases and 10.6% for more challenging scenarios (i.e. presence of intraparenchymal cysts, cortical defects). Lesion segmentation exhibited higher variability, with absorbed dose variability reaching up to 22.4%.

Segmentation significantly contributes to variability in absorbed dose estimates, particularly for lesions and for kidneys with anatomical complexities. Standardizing segmentation protocols and providing training on advanced segmentation methods are essential to reduce variability.