FDT-biotech: theranostic digital twins for personalized cancer treatment

The National Science Foundation Division of Mathematical Sciences awarded the University of Massachusetts $558,159 on November 1, 2026, under the Mathematical and Physical Sciences program (CFDA 47.049) to develop a theranostic digital twin computational model for personalized radiopharmaceutical therapy dosing in cancer treatment.

The project will create a continuously updated virtual patient model built from imaging scans and blood biomarkers collected before and during treatment. The digital twin combines mathematical equations describing drug transport through the body with artificial intelligence methods trained on patient data to forecast radiation dose distribution to tumors and healthy organs before each treatment cycle and recommend patient-tailored injection dosages. The framework addresses current clinical limitations in radiopharmaceutical therapy, where fixed dosages regardless of patient variation in body size, tumor burden, or organ clearance rates leave some patients undertreated while exposing others to organ toxicity. The mathematical framework applies to any radiopharmaceutical drug that can be imaged and delivered over multiple treatment cycles, with potential extension to cancers including neuroendocrine tumors.

Work is performed in Amherst, Massachusetts, with a period of performance from November 1, 2026, through October 31, 2029.

Abstract

Radiopharmaceutical therapy (RPT) is a new targeted approach to treating cancer that carries radiation directly to cancer cells while sparing the patient?s healthy organs. The RPT drug pluvicto, approved by the FDA in 2022 for men with metastatic prostate cancer, is now a widely used prostate cancer treatment shown to extend patients? Lives. Every patient today, however, receives the same, fixed RPT injection dosage regardless of variations in body size, tumor burden, or how quickly that person?s organs clear the drug. Published data show that this one size fits all approach leaves some patients undertreated while placing others at risk of kidney, liver, or bone marrow damage. No tool currently exists that can take the medical images and blood tests already collected during treatment, turn them into a personalized forecast of how much radiation a given patient?s tumor and organs will receive, and recommend patient-tailored treatment regimens. This project will address this unmet clinical need by developing a computational model called a theranostic digital twin, a continuously updated, virtual copy of each patient built from imaging scans and blood biomarkers collected before and during treatment. Inside the digital twin, mathematical equations describing how the drug moves through the body will be combined with artificial intelligence methods that learn from patient data, allowing the model to forecast before each treatment cycle how the radiation dose will distribute inside that person?s body and what its benefit and risk will be. IT will use the forecasts to choose an injection dosage for the specific patient to reduce tumor burden while keeping healthy organs safe. The mathematical framework built here applies to any RPT drug that can be imaged and delivered over multiple treatment cycles. IT is expected to extend to other cancers now being treated with RPT including neuroendocrine tumors. The project will train a doctoral student in this emerging field, integrate the work into two graduate and undergraduate university courses, and bring hands on learning activities to k-12 students in western massachusetts. This interdisciplinary project will design and validate a theranostic digital twin (TDT) for 177LU-psma radiopharmaceutical therapy (RPT) in metastatic prostate cancer. The TDT is a patient-specific computational model constructed as a physics-informed world model. IT couples a physiologically based pharmacokinetic differential equation system with a neural network. The TDT will be initialized from pre-therapy positron emission tomography (pet) and computed tomography (CT) imaging and updated in each treatment cycle with data from post-therapy single photon emission computed tomography (spect) imaging, injected activity and timing, and blood biomarkers. In the first objective, the investigators will construct and train the world model on a DE-identified, pre-existing patient dataset and forecast next-cycle pharmacokinetics, and organ- and tumor-absorbed dose with quantified uncertainty. In the second objective, the investigators will build decoder modules to forecast next-cycle tumor efficacy and renal, hepatic, and hematologic laboratory measures, and derive personalized recommendations for injected activity and inter-cycle timing from the model?s dose forecasts, evaluated by counterfactual simulation with alternative regimens. In the third objective, the trained model will be applied without retraining to an independent cohort and different imaging scanners to test generalizability. A cycle-by-cycle verification, validation, and uncertainty quantification framework will track prediction uncertainty as treatment progresses. All code (phantom generation, population pharmacokinetics, and model training) will be released under an open-source license. The completed framework will be submitted to the FDA office of science and engineering laboratories as a candidate regulatory science tool. The project contributes methods in physics-constrained modeling, uncertainty propagation through coupled mechanistic and neural differential equation models, and a continual verification and validation methodology for digital twins in the clinic. The methods could be extensible to other approved or investigational RPT drugs such as 177LU-dotatate for neuroendocrine tumors and alpha particle therapies. This project is jointly supported by the national institutes of health through the office of data science strategy and the national institute of biomedical imaging and bioengineering. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the foundation's intellectual merit and broader impacts review criteria.- subawards are not planned for this award.