Radiopharmaceutical therapy (RPT) has made great strides over the last 50 years, having only recently received FDA approval to treat prostate cancer. But despite the progress—and billions of dollars of industry investments—one area that has lagged behind is dosing, which is still one-size-fits-all. Now, researchers at the University of Massachusetts Amherst have developed an AI model that can inform treatment individualization by generating a patient’s radiation dose map that matches gold-standard accuracy in under 23 seconds. 

“Right now, everybody gets the same dose,” says Joyita Dutta, professor in the Riccio College of Engineering at UMass Amherst. “That essentially leaves the therapy’s potential untapped, to the extent that it’s suboptimal for a given patient. Measuring how much radiation each tissue actually absorbs is the key to personalizing treatment.”

RPT received FDA approval for treating late-stage prostate cancer in 2022. One of the most common cancers in men in the U.S. (second only to skin cancer), prostate cancer affects one in eight men in their lifetime. While survival rates are good if caught early, those whose cancer is diagnosed after it has spread have a five-year survival rate of only 40%.

Unlike traditional, localized radiation, RPT is injected. This allows it to travel through the whole body while still selectively targeting tumors. “The pharma companies realize the value of this,” says Dutta. “There’s been a series of acquisitions. Every big pharma company you can think of has acquired some radiopharmaceutical therapy startup developing new therapeutics or building technology around them.”

Yet, despite the industry attention and treatment advancements, individualized dosing has remained elusive. 

“The main issue with many cancer treatments is toxicity,” says Dutta. “Whether it’s radiation, chemo, or radiopharmaceutical therapy—whatever mechanism kills the cancer cells also hurts healthy tissue.” 

In theory, when doctors can see how much radiation is absorbed across the body after treatment (called dosimetry), they can determine if the patient can safely tolerate a higher dose in subsequent treatments, without risk to other organs. With RPT, post-treatment scans can reveal where the drug concentrates. “But the image by itself isn’t telling you what the dose is,” says Dutta. 

The gold-standard computational method for measuring radiation uptake is accurate but can take hours per patient, making it clinically impractical. This is where Dutta saw an opportunity to apply her lab’s expertise at creating AI-based biomedical tools. 

The team created DiffuDose, a combination of two AI modules, one that generates a coarse dose estimate and another that sharpens it into a full-resolution radiation dose map, matching gold-standard accuracy in under 23 seconds per patient. Against six competing methods, it achieved the best overall quantitative performance and had consistently high performance across multiple organs, including both kidneys and the liver, some of the key organs at risk for toxicity. Their work is published in IEEE Transactions on Radiation and Plasma Medical Sciences. 

“Pixel by pixel in a full image, you could see how the dose was distributed across the body,” says Dutta. “That’s what really helps you personalize the treatment,” both informing the quantity of drug administered and the frequency of treatments. 

Next, Dutta says she has an emerging collaboration with UMass Chan Medical School to continue to build AI models with post-RPT scans as well as patient blood biomarkers to get a better sense of how a patient is responding to the therapy. 

The work drew on contributions from collaborators outside UMass Amherst: Michael King at UMass Chan Medical School, Alejandro Bertolet and Quanzheng Li at Massachusetts General Hospital, and Babak Saboury at the Institute of Nuclear Medicine in Bethesda, Maryland. 

Dutta also highlights the achievements of her graduate student co-authors. Bowen Lei, a Ph.D. student and the first author on the paper, is a Young Investigator Competition candidate for his work in dosimetry at the 2026 IEEE Medical Imaging Conference to be held in Granada, Spain. 

Co-author Vibha Balaji, a Ph.D. student, received first place in the Physics, Instrumentation and Data Sciences (PIDS) Best Poster Award competition at the Society of Nuclear Medicine and Molecular Imaging (SNMMI) 2025 Annual Meeting in New Orleans, LA, for her radiopharmaceutical therapy work. 

Dutta also credits co-author Tzu-An Song, a postdoctoral researcher in her lab who is co-leading a translational effort on radiopharmaceutical therapy with her through a 2025 UMass Translational Seed Award, and Ziyuan Zhou, a graduate student in her lab who received a 2026-2027 UMass IALS Translational Graduate Student Assistantship to create novel image enhancement methods for radiopharmaceutical therapy.

Top image: A medical linear accelerator used in therapeutic oncology. Credit: Getty Images.