ETD

• dosimetry
• glands
• kidneys
• liver
• lutetium-177
• organs at risk
• radioligand therapy
• red marrow
• single-time-point dosimetry
• spleen

Prostate cancer is one of the most frequently diagnosed malignancies in men and remains a major cause of cancer-related mortality worldwide. In patients with metastatic castration-resistant prostate cancer (mCRPC), radioligand therapy (RLT) targeting the Prostate-Specific Membrane Antigen (PSMA) with $^{177}$Lu-PSMA has emerged as a promising treatment option. By targeting the PSMA, a transmembrane glycoprotein markedly overexpressed on the surface of prostate cancer cells, the radiopharmaceutical enables selective delivery of ionizing radiation to tumor lesions while limiting exposure to healthy tissues. The recent clinical introduction of $^{177}$Lu-PSMA, officially approved in Italy in March 2025, has increased the need for efficient dosimetric approaches capable of supporting treatment optimization. Personalized dosimetry plays a crucial role in evaluating absorbed doses to organs at risk (OARs), maximizing therapeutic efficacy and minimizing toxicity. However, conventional dosimetric protocols are often difficult to implement in routine clinical practice, as they typically require multiple imaging time points and, for red marrow (RM) dosimetry, repeated blood sampling. Furthermore, many historical approaches rely on planar imaging and external whole body measurements, which provide lower quantitative accuracy than modern three-dimensional tomographic techniques.
The aim of this work was to develop and validate a simplified and highly automated dosimetric workflow for patients undergoing $^{177}$Lu-PSMA therapy, based exclusively on the use of single-photon emission computed tomography/computed tomography (SPECT/CT) images, each acquired approximately 96 hours after radiopharmaceutical administration. In particular, the nnUNet-based TotalSegmentator framework, a fully automatic segmentation algorithm, was introduced and validated in clinical practice for the delineation of large OARs on CT images. In addition, the study extended dosimetric analysis to small structures, such as salivary and lacrimal glands, whose dosimetry is particularly challenging due to partial volume effects and has only recently been introduced into clinical practice. Finally, a novel method for RM dosimetry was proposed in this thesis.
The study included 35 patients treated according to the standard $^{177}$Lu-PSMA protocol, consisting of up to six cycles of 7.4 GBq, for a total of 94 analyzed treatment cycles. Dosimetry was based exclusively on quantitative SPECT/CT images acquired approximately 96 hours post injection. For large organs, including kidneys, liver and spleen, segmentation was performed automatically using TotalSegmentator. This approach enabled rapid, reproducible and operator-independent organ delineation. Absorbed dose estimation for these organs was performed using the H\"anscheid single-time-point method. The same single-time-point approach was also applied to small structures, including the salivary and lacrimal glands. Their segmentation was performed using MIM Contour ProtégéAI, a deep-learning-based auto-segmentation tool whose neural network model is based on the U-Net architecture. To improve quantitative accuracy for these small structures, partial volume effect correction was applied through recovery coefficients derived from experimental measurements performed with a NEMA IEC phantom. For RM dosimetry, a comprehensive dosimetric assessment was performed through a single post-therapy imaging session, avoiding traditional approaches based on multiple planar acquisitions or serial blood sampling. The RM dosimetry method accounted for three dose contributions: self-dose, cross-dose and remainder-body contributions. The self-dose component was estimated from activity measured in the L2-L4 vertebrae segmented with TotalSegmentator and subsequently scaled to the entire RM compartment using anatomical reference models. This analysis was restricted to 8 patients without metastatic involvement of the selected vertebrae.
One of the main innovations of the proposed workflow is therefore the exclusive use of quantitative three-dimensional imaging for the dosimetric assessment of large organs, small glands and RM. This significantly reduces patient burden, an important advantage considering the advanced clinical condition of many mCRPC patients, while simultaneously optimizing hospital resources and improving workflow sustainability.
The absorbed dose estimates for kidneys, liver, spleen and RM were overall comparable with those reported in the literature. Considerable inter-patient variability was observed across the investigated organs, with individual dose values spanning wide ranges, highlighting the importance of patient-specific dosimetric assessment. Similarly, the absorbed dose estimates for the salivary and lacrimal glands were consistent with the available literature, although substantial inter-patient variability was observed. Despite this variability, the estimated dose ranges remained below the reference limits reported in the literature and adopted in this study, supporting the applicability of the proposed approach for patient-specific dosimetric assessment.
Overall, the proposed workflow proved to be feasible, and compatible with routine clinical implementation. The combination of quantitative 3D imaging, automated segmentation through nnUNet-based tools, dedicated analysis of small glands and single-time-point dosimetry represents a promising strategy for personalized dosimetric assessment in $^{177}$Lu-PSMA therapy. Improvements include extending the methodology to patients with metastatic involvement of the L2-L4 vertebrae and introducing patient-specific effective half-life estimates for individual organs, replacing literature-based average values. The ultimate goal is to establish a rapid, fully automated, and patient-tailored dosimetric workflow capable of supporting personalized dose-escalation strategies, once data from a larger cohort of patients completing all six therapy cycles become available. Such an approach could increase treatment efficacy by maximizing tumor absorbed dose while preserving OARs, thereby contributing to the development of increasingly personalized and optimized radioligand therapies.