ETD

• Lu-177-DOTATATE

Radioligand Therapy (RLT) with $^{177}$Lu-DOTATATE is an established treatment for advanced neuroendocrine tumors because of high affinity of $^{177}$Lu for somatostatin receptors (SSTRs) by tumor cells. However, renal clearance exposes the kidney to considerable radiation doses, making it the dose-limiting organ. The current clinical protocol consists of administration of a fixed activity of 7,4 GBq per cycle to all patients, regardless of individual characteristics. Hence, the lack of personalized dosimetry limits optimizing treatment to maximize therapeutic effect while minimizing renal toxicity risk.\\
This work aims to develop and validate an optimized renal dosimetry workflow for patients undergoing RLT with $^{177}$Lu-DOTATATE to support future customization of the activity administered in cycles following the first.\\
A cohort of 74 patients was analyzed, for a total of 251 treatment cycles. SPECT/CT acquisitions were performed following a standardized clinical protocol and calibrated by phantom measurements, comparing different software tools (Python, LIFEx, Slicer3D). Renal segmentation was fully automatic with the TotalSegmentator deep learning algorithm (nn-UNet architecture) performed on CT images from hybrid SPECT/CT. Dosimetry was calculated using the MIRD formalism and the H\"anscheid single-time point method based on a single acquisition 96 hours after administration, reducing the effort required to patients. Subsequently, the mean renal absorbed dose was converted to a biologically effective dose (BED) using the linear-quadratic model. To predict total cumulative dose after four therapy cycles, machine learning models (linear parametric regression, Random Forest, Extra Trees, and Boosting) were implemented using 5-fold cross-validation and compared, evaluating performance with R², MSE, and RMSE. Based on the predicted cumulative renal dose, a personalized activity planning analysis was performed to estimate the maximum activity that could be administered in subsequent cycles while remaining within established dosimetric safety thresholds, allowing for optimization of patient-specific treatment.\\
The results showed good renal dose stability between treatment cycles, confirming therapy reproducibility, and notable inter-patient variability, with some cases approaching or exceeding safety limits reported for Peptide Receptor Radionuclide Therapy (PRRT). Analysis revealed that approximately $59\%$ of patients remained within safe dose limits, while the remaining $41\%$ approached or exceeded established safety thresholds. The best-performing predictive model was linear regression based on first-cycle dose ($R^2=0,90\pm0,3$), and also Extra Trees ($R^2=0,84\pm0,11$), evaluated on the test set.\\
In conclusion, this work allowed for accurate and clinically sustainable monitoring of renal exposure over four therapeutic cycles. Integrating a predictive model based on first-cycle data produces a tool that allows RLT to be customized with $^{177}$Lu-DOTATATE, allowing modulation of administered activity within recommended dosimetric limits to maximize therapeutic efficacy with an adequate safety profile.