UMass AI Model Personalizes Prostate Cancer Dosimetry

Researchers at the University of Massachusetts Amherst have developed DiffuDose, an AI model that generates personalized radiation dose maps for radiopharmaceutical therapy (RPT) in prostate cancer patients. The model matches gold-standard accuracy in under 23 seconds, compared to hours for conventional methods. This advancement addresses the one-size-fits-all dosing limitation of RPT, which received FDA approval for late-stage prostate cancer in 2022. The model uses diffusion-guided deep learning to produce pixel-by-pixel dose distributions, enabling clinicians to tailor treatment doses and frequencies based on individual patient radiation absorption. In testing against six competing methods, DiffuDose achieved the best overall performance, particularly in organs at risk for toxicity such as kidneys and liver. The research was published in Institute of Electrical and Electronics Engineers Transactions on Radiation and Plasma Medical Sciences. The team is collaborating with UMass Chan Medical School to further develop models incorporating blood biomarkers. This breakthrough could transform RPT from a standardized treatment to a more personalized therapy, potentially improving outcomes and reducing toxicity for prostate cancer patients.