A novel machine-learning methodology designed for the treatment of metastatic castration-resistant prostate cancer (mCRPC) utilizing prostate-specific membrane antigen (PSMA) has demonstrated the capacity to forecast radiation doses administered to neoplastic growths and healthy tissues prior to therapeutic intervention. By leveraging data readily obtainable from pre-treatment PET/CT scans, this innovative predictive instrument holds the potential to facilitate tailored treatment regimens, enhance patient selection processes, and mitigate the risks associated with treatment-induced toxicity. This groundbreaking research was formally presented at the Society of Nuclear Medicine and Molecular Imaging’s 2026 Annual Meeting.
The precise estimation of radiation dose, known as dosimetry, is an indispensable component for the optimization of ⁷⁷Lu-PSMA radiopharmaceutical therapy in cases of mCRPC. Currently, the standard practice involves utilizing post-therapy imaging to compute dosimetry; however, this process is both time-consuming and demands significant resources. Conversely, pre-therapy PET/CT scans present a valuable opportunity to evaluate the anticipated efficacy of treatment and potential risks before the commencement of therapy.
The routine administration and widespread availability of 18F-PSMA PET/CT scans in patients diagnosed with prostate cancer have been established; however, their utility in forecasting therapeutic radiation doses had not been previously investigated. Our investigation was specifically designed to ascertain whether the information inherent in these scans could inform treatment planning prior to therapy initiation and thereby foster a more individualized approach to patient care.”
Amit Nautiyal, PhD, a scientist and National Institute for Health and Care Research (NIHR) fellow at University Hospital Southampton and the University of Southampton, United Kingdom
This initial proof-of-concept investigation enrolled nine patients afflicted with mCRPC who were candidates for ⁷⁷Lu-PSMA radiopharmaceutical therapy. The dataset comprised analyses from 57 tumors, 36 salivary glands, and 18 kidneys. Researchers meticulously developed a machine learning mixed effects model with the objective of predicting absorbed radiation doses within both tumors and organs. The predictive factors incorporated uptake-based PET metrics, radiomic characteristics, and relevant clinical biomarkers. The accuracy of the predictive estimates was subsequently substantiated through a comparison with dosimetry calculations derived from imaging obtained after one cycle of ⁷⁷Lu-PSMA therapy.
The machine learning model, which is predicated on pre-therapy 18F-PSMA PET/CT scans, exhibited a notable proficiency in forecasting the absorbed radiation doses to tumors and vital organs. By integrating uptake features, radiomic data, and clinical biomarkers, while simultaneously accounting for patient-specific variations, the model demonstrates considerable promise in utilizing pre-therapeutic diagnostic information to predict post-therapeutic dosimetry.
“Should this methodology be validated through more extensive studies, it could lead to enhanced patient selection and more informed decision-making during the pre-treatment evaluation phase, ultimately contributing to the optimization of ⁷⁷Lu-PSMA therapy on an individual patient basis. On a broader scale, this research underscores the evolving role of medical imaging, moving beyond mere diagnostic capabilities to actively guiding personalized treatment strategies,” remarked Nautiyal.
This foundational proof-of-concept research forms an integral part of a projected five-year initiative dedicated to the accumulation of further data and the subsequent development of a robust and validated predictive model. This endeavor has received vital financial support from the NIHR in the United Kingdom. Future research efforts will be directed towards large-scale, multi-center cohort studies to refine the accuracy of pre-therapy absorbed dose predictions and to conduct independent validation studies. These undertakings will ultimately support the effective stratification of patients for personalized ⁷⁷Lu-PSMA radiopharmaceutical therapy within the realm of clinical practice.
