Sensitivity of Dosimetry Calculations to Tumor Segmentation Methods on SPECT/CT Images for Patients Undergoing 177 Lu-PSMA-617 Radiopharmaceutical Therapy (RPT)
Abstract
Purpose
Radiopharmaceutical therapy (RPT) delivers targeted radiation to cancer cells. Patient‑specific dosimetry can help estimate dose and predict treatment outcome. However, tumor segmentation approaches vary across institutions, introducing uncertainty into dose estimates and potentially impacting clinical decision making. This study evaluates the sensitivity of RPT dosimetry to differences in tumor segmentation techniques
Methods
Quantitative SPECT/CT scans from 12 patients were analyzed at 1 hour (TP1), 24 hours (TP2), and 96 hours (TP3) after administration of 200 mCi of 177Lu‑PSMA‑617. At each timepoint, tumors were segmented using local thresholding with SUV>2.5 and a gradient‑based method (PETEdge+, PE+). Segmentation results were merged to form a unified “tumor_all” contour. For 5 patients, voxel‑based dosimetry was performed using time‑activity curves from all SPECT/CT timepoints and Monte Carlo dose calculations in Torch (Voximetry). Contours were compared using the Dice similarity coefficient and tumor volume normalized to physician‑drawn tumor contours. Dose comparisons were evaluated using mean dose to tumor_all normalized to the physician‑derived mean dose. All comparisons were assessed across segmentation methods and timepoints.
Results
SUV thresholding consistently produced smaller tumor volumes than PE+, particularly at TP1 (P<0.001) and TP2 (P=0.04). Across timepoints, PE+ achieved higher agreement (P<0.001) with physician contours (Dice: 0.64±0.10) compared with SUV thresholding (Dice: 0.52±0.21). PE+ at TP3 showed the strongest overall alignment (Dice: 0.71±0.12). Dosimetry analysis revealed that PE+ at TP2 produced mean tumor doses closest to physician contours (Normalized Mean dose: 0.97±0.34). SUV thresholding led to dose overestimation of nearly 50% at all three timepoints.
Conclusion
Segmentation method and imaging timepoint substantially influence tumor delineation and downstream RPT dose calculations. PE+ at TP3 best approximated physician contours, while PE+ at TP2 yielded the most accurate dose estimates. These findings underscore the importance of standardized segmentation approaches to improve the reliability of patient‑specific RPT dosimetry.