Dosimetry—the precise measurement of absorbed radiation dose—remains an indispensable component in refining and optimizing radionuclide therapies such as ^177Lu-PSMA. Traditionally, dosimetric evaluation relies heavily on imaging conducted post-treatment, which poses significant challenges due to its labor-intensive nature and the extensive resources required. The advent of a pre-therapy predictive tool utilizing widely available ^18F-PSMA PET/CT imaging represents a major leap forward by potentially circumventing these constraints. This shift not only promises to streamline clinical workflows but also extends the possibility of tailoring treatment intensity to individual patient profiles, thus maximizing therapeutic benefit while minimizing adverse effects.

The research, spearheaded by Dr. Amit Nautiyal and colleagues at the University Hospital Southampton and the University of Southampton, UK, employs a sophisticated machine learning framework combining mixed-effects modeling with multi-parametric data inputs. The model assimilates PET uptake metrics, radiomic features—which capture spatial and textural heterogeneity of lesions—and relevant clinical biomarkers. By integrating these multidimensional variables, the algorithm can accommodate inter-patient variability and predict absorbed dose distributions in tumors alongside vital organs such as salivary glands and kidneys with promising accuracy.

This proof-of-concept study analyzed data from nine mCRPC patients undergoing ^177Lu-PSMA therapy. Across these individuals, 57 tumors, 36 salivary glands, and 18 kidneys were evaluated, offering a robust dataset for model training and validation. The comparison of predicted absorbed doses with those calculated via conventional post-therapy imaging demonstrated the model’s potential in accurately forecasting dosimetric outcomes prior to treatment initiation. Such validation underscores how comprehensive image-derived quantitative features, when harnessed through machine learning techniques, can revolutionize personalized treatment planning in nuclear medicine.

One of the critical advantages of this approach lies in its capacity to inform patient selection. By predicting which patients are likely to receive optimal radiation doses in tumors while sparing normal tissue, clinicians can better stratify candidates for ^177Lu-PSMA therapy. This strategic selection inherently reduces the risk of treatment-associated toxicity and enhances the likelihood of favorable clinical responses. Furthermore, this predictive capacity may serve as an invaluable decision support tool during multidisciplinary team discussions, where tailored therapeutic regimens are formulated based on individual risk-benefit assessments.

The integration of radiomics—a burgeoning field that quantitatively analyzes medical images beyond conventional visual interpretation—marks a significant step forward in nuclear oncology. The nuanced information extracted from texture, shape, and intensity patterns within the ^18F-PSMA PET/CT images provides a rich dataset that machine learning algorithms can exploit to uncover complex relationships correlating with dosimetric parameters. When combined with patient-specific clinical biomarkers, this multifaceted modeling embodies the essence of precision medicine, ensuring treatment is dynamically adapted to each patient’s unique biological landscape.

Dr. Nautiyal emphasizes the transformative potential of this methodology, suggesting that, pending corroboration through larger cohort studies, it could redefine pre-treatment assessment strategies globally. Such validation would not only affirm the reproducibility and scalability of the model but also encourage its adoption into routine clinical practice. The ability to anticipate radiation dose distributions before therapy confers tangible benefits, including reduced need for extensive post-therapy imaging, diminished patient burden, and expedited initiation of treatment cycles.

The current research represents a foundational step in a comprehensive five-year initiative aimed at expanding the training dataset, refining the predictive accuracy of the model, and conducting rigorous external validation using multi-center patient cohorts. This longitudinal program aspires to establish a robust, clinically deployable tool capable of stratifying patients effectively and personalizing ^177Lu-PSMA radiopharmaceutical therapy. Importantly, the ongoing collaboration across institutions highlights the multidisciplinary nature of this endeavor, spanning nuclear medicine, radiology, oncology, and data science.

From a technical perspective, the employment of mixed-effects models within the machine learning framework allows for the accommodation of both fixed effects related to PET and clinical features and random effects capturing patient-specific variabilities. This statistical architecture enhances the model’s flexibility and adaptability across heterogeneous patient populations, which is paramount given the variability inherent in tumor biology and organ susceptibility. It also mitigates potential biases that might arise from limited sample sizes, fostering generalizability.

The implications of this work extend beyond prostate cancer and ^177Lu-PSMA therapy. The demonstrated feasibility of using pre-treatment imaging combined with advanced computational analytics to predict treatment dosimetry could inspire similar approaches across various theranostic applications. This positions imaging not merely as a diagnostic modality but as a dynamic, integral component of personalized therapy planning, bridging the gap between molecular visualization and actionable clinical insights.

In conclusion, this compelling study from the University of Southampton consortium delivers a visionary framework for enhancing the precision and efficacy of radionuclide therapy in advanced prostate cancer. By harnessing routinely acquired ^18F-PSMA PET/CT data through machine learning innovation, the research charts a path toward individualized treatment strategies that promise to improve patient outcomes significantly. As this technology progresses toward clinical translation, it heralds a paradigm shift in nuclear medicine, where therapy is foreseen and optimized well before a radioactive agent is administered.

Subject of Research: Machine learning for pre-therapy prediction of tumor and organ absorbed dose in ^177Lu-PSMA radiopharmaceutical therapy using ^18F-PSMA PET/CT radiomics and clinical biomarkers.

Article Title: Machine Learning-Based Pretherapy Prediction of Tumor and Organ Absorbed Dose in ^177Lu-PSMA Therapy Using ^18F-PSMA PET/CT Radiomics and Biomarkers

News Publication Date: 2026 (presented at SNMMI 2026 Annual Meeting)

Web References:

• Link to Abstract
• Society of Nuclear Medicine and Molecular Imaging official site: snmmi.org

References:

• Nautiyal A., Crabb S., Martinez Camacho R., Sundram F., Saad Z., Michopoulou S., Dewaraja Y., Dickson J. Machine Learning-Based Pretherapy Prediction of Tumour and Organ Absorbed Dose in ^177Lu-PSMA Therapy Using ^18F-PSMA PET/CT Radiomics and Biomarkers. SNMMI 2026 Annual Meeting, Abstract 262138.

Image Credits: Courtesy of SNMMI

Keywords: molecular imaging, positron emission tomography, radiopharmaceutical therapy, prostate cancer, ^177Lu-PSMA therapy, ^18F-PSMA PET/CT, dosimetry, machine learning, radiomics, personalized medicine, metastatic castration-resistant prostate cancer, nuclear medicine