A new machine learning approach to prostate-specific membrane antigen (PSMA) treatment of metastatic castration-resistant prostate cancer (mCRPC) could estimate radiation dose to tumors and healthy organs before starting treatment. Using data already available from pre-treatment PET/CT scans, this new predictive tool could help personalize treatment plans, improve patient selection and reduce the risk of toxicity. This research was presented at the 2026 Society of Nuclear Medicine and Molecular Imaging Annual Meeting.
Dosimetry is critical for optimizing 77Lu-PSMA radiopharmaceutical therapy in mCRPC. Currently, post-treatment imaging is commonly used to calculate dosimetry. However, it is time consuming and resource intensive. Pretreatment PET/CT offers an opportunity to assess the potential efficacy and risk of treatment prior to treatment.
18F-PSMA PET/CT is already routinely performed and widely available in prostate cancer patients, but its ability to predict treatment radiation dose has not been previously investigated. Our study sought to determine whether the already available information from these scans could guide pre-treatment planning and support more personalized care.”
Amit Nautiyal, PhD, Scientist and National Institute for Health and Care Research (NIHR) Fellow at University Hospital Southampton and University of Southampton, UK
In this proof-of-concept study, nine mCRPC patients referred for 77Lu-PSMA radiopharmaceutical therapy contributed 57 tumors, 36 salivary glands, and 18 kidneys for analysis. The researchers developed a machine learning mixed-effects model to predict absorbed doses to tumors and organs. Prognostic factors included uptake-based PET measurements, radiological features and clinical biomarkers. Predictive estimates were compared with dosimetry calculated after one cycle of 77Lu-PSMA treatment to assess accuracy.
The pre-treatment 18The machine learning model based on PET/CT F-PSMA showed a promising ability to predict tumor and organ absorbed dose. By combining uptake characteristics, radioactivity and clinical biomarkers while accounting for patient-level variability, the model shows potential for using pre-treatment information to predict post-treatment dosimetry.
“If validated in larger studies, this approach may improve patient selection and support better decision-making during pre-treatment assessment, helping to optimize 77Lu-PSMA therapy for individual patients. More generally, it highlights how imaging can go beyond diagnosis and actively guide personalized therapy,” said Nautiyal.
This proof-of-concept research is part of a planned five-year program to collect more data and develop a robust, validated model. This work was supported by the NIHR in the UK. Future work will focus on larger, multicenter cohorts to improve pretreatment absorbed dose predictions and perform independent validation to support patient stratification for personalized 77Lu-PSMA radiopharmaceutical therapy in clinical practice.
