New AI-based Technologies for Even Safer and More Precise Nuclear Medicine

This is a descriptive, observational, non-profit study aimed at detecting and predicting extravasation events during the administration of radiopharmaceuticals for diagnostic and therapeutic purposes in nuclear medicine. Extravasation can lead to local tissue damage and compromise the accuracy of semi-quantitative imaging parameters such as the Standardized Uptake Value (SUV), widely used in PET/CT for diagnosis, staging, and therapy response evaluation. Literature reports that extravasation may cause a 21-50% change in SUV, potentially leading to incorrect assessment of tumor response.

The study will use a CE-marked portable spectroscopic personal radiation detector (RadEye SPRD-ER, Thermo Fisher Scientific™), already validated in a previous Ethics Committee-approved study, to record dose-rate (DR) curves during radiopharmaceutical injections. Using these data, new dosimetric metrics will be developed to characterize correct, abnormal, and extravasation events. Machine learning (ML) algorithms will be trained on patient clinical data, injection metrics, and DR curves to classify injection events in real time and to estimate correction factors for SUV quantification. Monte Carlo simulations (MCNP code, anthropomorphic phantoms, and reconstructed patient geometries) will be performed to evaluate absorbed dose distributions in extravascular regions.

The project is structured into three phases:

Phase 1 (Data Acquisition & Analysis): Real-time monitoring with RadEye SPRD-ER, extraction of quantitative metrics (DRmax, DRmean, Δp, t*, Δt), development of ML classifiers and regression models for SUV correction.

Phase 2 (Monte Carlo Simulations): Activity and dose calibration, dose distribution modeling in extravascular tissues.