Computational Nuclear Oncology for Precision Radiopharmaceutical Therapies

Introduction: Radiopharmaceutical therapy (RPT), with its targeted delivery of cytotoxic ionizing radiation, demonstrates significant potential value for patients with metastatic disease. Currently, positron emission tomography (PET) imaging is used to determine RPT suitability with fixed therapeutic activity administration. There is an opportunity to optimize RPTs and enhance the precision of theranostics by moving beyond a ‘one-size-fits-all’ approach and using patient-specific image-based dosimetry for personalized treatment planning. Such an approach, however, requires accurate prediction of clinical outcomes before treatment planning can be performed. Identification of suitably predictive biomarkers of outcomes requires advanced computational tools and standardization of both data collection and modeling protocols in order to enable reliability and generalizability. Robust and flexible approaches for data and model integration are necessary to this end.

Methods: The SNMMI AI-Dosimetry Working Group is actively developing and expanding the field of "computational nuclear oncology (CNO)", a unified set of scientific principles and mathematical models that describe the hierarchy of etiological mechanisms that are involved in RPT dose-response. This includes ‘upstream’ radio-pharmacokinetics, ‘midstream’ dosimetry and radiobiology, and ‘downstream’ (patho-)physiological response leading to clinical endpoints. These explicit ‘bottom-up’ mechanism-based models, and their combination with ‘top-down’ data-driven models, are outlined, with specific reference to explainable artificial intelligence and interpretable machine learning. Particular focus is provided to efforts to integrate disparate tools through the lens of harmonizing heterogenous sources of information. In addition, major social, ethical and regulatory challenges that may arise from the data collection, integration, and modeling process are discussed.

Results: CNO serves as a common ‘generative model’ for computations and datasets in RPT. In the short- and intermediate-term, CNO is envisioned to serve as a universal semantic ontological framework that facilitates information exchange and collaborative modeling across the nuclear medicine community. In doing so, it can catalyze a collective transition to dosimetry-driven personalized RPT. In the long-term, CNO is anticipated to support a framework for interpretability and modularity with the ancillary benefit of enabling nuclear medicine to effectively communicate with other communities across the larger ‘AI ecosystem’, particularly those in radiation and clinical oncology, radiation epidemiology and protection, and preclinical radiation sciences.

Conclusions: The overview of tools available in CNO illustrated a mechanism-based framework consistent with the journey towards personalized dosimetry-driven RPT. Facilitation of the information exchange necessary to identify robust dosimetric correlates of outcomes that can be used to perform patient-specific treatment optimization were highlighted. While there are myriad benefits and opportunities, implementation of this technology in a socially responsible manner will require special care and consideration.

• Download figure
• Open in new tab
• Download powerpoint