An AI Model Just Made Dosing Guesswork Obsolete

UMass Amherst's DiffuDose matches gold-standard accuracy while personalizing dose maps in under 23 seconds.

12 Aug 2026

Researchers at the University of Massachusetts Amherst have built an artificial intelligence model that generates a patient specific radiation dose map with gold standard accuracy in under 23 seconds. The tool, named DiffuDose, targets one of radiopharmaceutical therapy's most persistent limitations: a one size fits all approach to dosing that has changed little even as the therapies themselves have advanced.

Radiopharmaceutical therapy won FDA approval to treat late stage prostate cancer in 2022, yet dosing has largely remained standardized across patients regardless of individual anatomy or organ sensitivity. Joyita Dutta, a professor in the Riccio College of Engineering at UMass Amherst, said uniform dosing leaves much of the therapy's potential untapped, since measuring how much radiation each tissue actually absorbs is the key to personalizing treatment.

DiffuDose uses a diffusion guided architecture to compute dose maps for patients receiving lutetium-177 PSMA radiopharmaceutical therapy, a process that traditionally required lengthy manual calculation. Compressing that work into seconds could let clinicians adjust treatment plans in near real time rather than relying on population level averages.

The research, published in IEEE Transactions on Radiation and Plasma Medical Sciences, credits contributions from doctoral students Bowen Lei, Vibha Balaji and Ziyuan Zhou alongside postdoctoral researcher Tzu-An Song. As AI driven dosimetry tools move from academic development toward clinical validation, they stand to complement the industry's broader push toward individualized radioligand therapy.