PET Scans Before Therapy Predict Radiation Doses in Prostate Cancer Treatment

Researchers have shown that a standard pre-therapy PET scan can predict how much radiation a patient’s tumors will absorb during a widely used prostate cancer treatment, potentially letting clinicians estimate dosimetry before the first treatment cycle. The finding comes from a study published in EJNMMI Physics that set out to predict absorbed dose to whole-body tumor volume from [177Lu]Lu-PSMA-617 radiopharmaceutical therapy using pre-therapy [68Ga]Ga-PSMA-11 PET metrics and common blood biomarkers.

Absorbed dose to whole-body tumor volume from [177Lu]Lu-PSMA-617 therapy has previously been shown to correlate with treatment response. Because of that link, the study’s authors reasoned that estimating this dose before therapy could help clinicians with patient selection and dosimetry-guided treatment planning. Their goal was to predict the dose delivered after the first treatment cycle using information already available at baseline.

The study included 31 patients. Each patient underwent a baseline [68Ga]Ga-PSMA-11 PET/CT scan before starting [177Lu]Lu-PSMA-617 therapy, and then 3 to 4 SPECT/CT scans were performed after cycle 1 to measure the actual absorbed dose. Whole-body tumor volume for each patient was segmented using a commercially available, semi-automated tool and verified by a radiologist. In addition, up to six index tumors per patient were segmented, while kidneys and salivary glands were outlined using deep-learning tools.

The researchers evaluated six PET metrics and 15 blood biomarkers as potential predictors. They built univariable and multivariable linear regression models using three different feature sets — biomarker-only, PET-only, and PET combined with biomarkers — as well as random forest regression machine-learning models that incorporated all features. The targets were cycle 1 absorbed dose to whole-body tumor volumes, index tumors, kidneys, and salivary glands.

Feature selection for the linear regression models used least absolute shrinkage and selection operator (LASSO) with a stability selection threshold of 0.7. Model performance was assessed with leave-one-patient-out cross-validation, reported as root mean squared error (RMSE) and median absolute percent error (median APE). Sensitivity analyses also tested random forest classification and logistic regression models for predicting binary prostate-specific antigen (PSA) response, defined as absorbed dose to whole-body tumor volume above previously identified thresholds.

The results showed that SUVMean was the only feature selected from the PET-only and PET-plus-biomarker feature sets for whole-body tumor volume. The resulting linear regression model achieved a cross-validated median absolute percent error of 42% and an RMSE of 0.69 Gy/GBq for predicting absorbed dose. The random forest model improved the median error to 27%, though with a slightly higher RMSE of 0.76 Gy/GBq.

For the other structures, linear regression models achieved similar but lower RMSE and median APE compared to the random forest models: for index tumors, an RMSE of 2.51 Gy/GBq and a median APE of 56%; for kidneys, 0.20 Gy/GBq and 32%; and for salivary glands, 0.16 Gy/GBq and 28%. Notably, blood biomarkers added little predictive value beyond the PET-derived metrics for the linear regression models.

The sensitivity analyses also pointed to useful classification performance. Sensitivity, specificity, and area under the receiver operating characteristic curve for predicting binary PSA response were high for both the random forest classification and logistic regression models, suggesting the pre-therapy data could also flag which patients are likely to reach dose thresholds associated with PSA response.

The authors conclude that cycle 1 [177Lu]Lu-PSMA-617 absorbed dose to whole-body tumor volumes, index tumors, kidneys, and salivary glands can be predicted using pre-therapy [68Ga]Ga-PSMA-11 PET metrics, and that across all structures the random forest and linear regression approaches provided comparable prediction performance. They caution, however, that before such models can be implemented clinically, the findings must be verified using larger and external data sets — an important limitation given the small, single-cohort size of 31 patients in this analysis.

Subject of Research: Predicting absorbed dose from lutetium-177 PSMA prostate cancer therapy using pre-therapy gallium-68 PSMA PET metrics and blood biomarkers

Article Title: Predicting [177Lu]Lu-PSMA-617 SPECT/CT-based absorbed dose to whole-body tumor volume from pre-therapy [68Ga]Ga-PSMA-11 PET metrics and common blood biomarkers

Article References: Fitzpatrick, K. J., Suresh, K., Roseland, M. E., Niedbala, J., Ngai, K., Wong, K. K., Frey, K. A., & Dewaraja, Y. K. (2026). Predicting [177Lu]Lu-PSMA-617 SPECT/CT-based absorbed dose to whole-body tumor volume from pre-therapy [68Ga]Ga-PSMA-11 PET metrics and common blood biomarkers. EJNMMI Physics. https://doi.org/10.1186/s40658-026-00939-4

Image Credits: AI Generated

DOI: 10.1186/s40658-026-00939-4

Keywords: radiopharmaceutical therapy, PSMA-617, PSMA-11 PET, prostate cancer, dosimetry, SPECT/CT, absorbed dose, machine learning, random forest, linear regression, biomarkers, PSA response