Development of artificial intelligence for simplified dosimetry in radiopharmaceutical therapy

Radiopharmaceutical therapy (RPT) using 177Lu-targeting radiopharmaceuticals has emerged as a promising treatment option for patients with metastatic castration-resistant prostate cancer (mCRPC). Accurate internal dosimetry is essential to quantify absorbed doses in tumour tissues and organs-at-risk (OAR), enabling optimization of therapy and prediction of clinical outcomes. However, conventional multiple-time-point (MTP) dosimetry requires several imaging sessions over multiple days, imposing logistical challenges, increased costs, and patient burden. As a result, simplified singletime point (STP) approaches have been proposed, though their accuracy and robustness remain a subject of ongoing investigation.
This thesis presents a comprehensive study on advancements in STP dosimetry and the integration of artificial intelligence methods for organ and tumor dosimetry in RPT. The first part provides an overview of mCRPC management, fundamental principles of internal dosimetry, and the current state of AI applications in nuclear medicine, including machine learning (ML) and support vector regressor (SVR) models. A systematic review of recent developments in STP dosimetry highlights methodological variations, performance trends, and clinical implications. Building on these findings, a novel instant STP (iSTP) dosimetry framework was developed using ML-based prediction of the effective half-life directly from early imaging data. The proposed model was trained and validated on clinical datasets, demonstrating satisfied correlation with reference MTP dosimetry and improved early and flexible time-point independence. Further analyses assessed model generalizability and robustness across patient cohorts, treatment cycles, and tumour burden levels.
The iSTP approach achieved accurate and reproducible dose estimates, significantly reducing the need for multi-time imaging without compromising precision. In conclusion, this work demonstrates the feasibility of AI-driven iSTP dosimetry for 177Lu-targeting radiopharmaceuticals in RPT, providing a foundation for faster, more patient-friendly, and standardized dosimetric workflows. These advancements may pave the way toward simplified dosimetry in radiopharmaceutical therapy, improving treatment efficacy and safety in mCRPC and beyond.

dissertation

true