arXiv:2510.21054v1 Announce Type: new
Abstract: Predictive dosimetry is central to enabling personalization of radiopharmaceutical therapies (RPTs) such as prostate-specific membrane antigen (PSMA) targeted RPTs. This study integrates physiologically based pharmacokinetic (PBPK) modeling with machine learning (ML) to predict physical (AUC, Dose) and biological (BED, EQD2) dosimetry in tumors and five organs. Using 640 realistically generated virtual patients, we simulated 15,360 time-activity curves (TACs) reflecting diverse uptake patterns. TAC-derived features trained ML models (RF, ET, Ridge, GB, XGBoost), with performance evaluated by Mean Absolute Percentage Error (MAPE). SHAP analysis identified key feature contributions varying by organ, endpoint, and tumor volume. Cu-64 based imaging yielded the most robust predictions, with dose prediction MAPE as low as 8% for tumors and 10-20% for different organs, while F-18 showed strong but more volume-dependent trends, and Ga-68 exhibited higher variability. The proposed PBPK-ML virtual theranostic trial framework enables robust predictive dosimetry and clinical trial design and optimization, advancing personalized planning for PSMA-targeted RPTs.