Development of model-based biophantoms for personalized cancer care

Introduction: We are developing model-based biophantoms for cancer imaging, treatment planning, and treatment response evaluation with the ultimate goal of providing personalized cancer care. Ongoing developments sharing a similar goal include the digital twins (DT) in clinical oncology and virtual clinical trials (VCT). DTs aim to predict and optimize treatment response via a patient-specific computational model (e.g., a tumor growth model), while VCTs aim to evaluate imaging performance, such as patient-specific protocol optimization, in silico by simulating the image formation chain using digital phantoms that mimic a patient population. Our biophantom project encompasses elements from both DTs and VCTs hence can be viewed as a hybrid between them. We develop image-derived tumor growth and metastasis models as in DTs, and subsequently incorporate the tumors into an existing phantom database that models the normal patient population. These bio-inspired digital phantoms are used as input to well-validated data generation pipelines to simulate clinical PET, SPECT, and CT images for imaging performance evaluation, treatment planning, and treatment response optimization and evaluation. This unique, end-to-end, mathematical and computational framework from image-derived tumor models to image-based treatment response optimization and evaluation characterizes the biophantom project.

Methods: The model-based biophantoms are built upon anatomically accurate computer-generated phantoms with various body shapes and organ sizes that model a normal patient population with additional enhancements to simulate tumor growth and distant metastasis. The "model" in model-based biophantoms refers to the physics model in the data generation pipeline, as well as image-derived anatomical and physiological models of tumor growth and pharmacokinetic models of the biodistribution of therapeutic agents. For our initial pilot study, we focus on PSMA-targeting agents used for metastatic prostate cancer imaging and treatment, but the mathematical and computational frameworks are generalizable to other agents and disease sites.

Results: Mesh models of prostate tumors are obtained from the Cancer Imaging Archive. The parametric tumor models are incorporated into a 3D mesh model of the XCAT phantom to derive 4D mesh models simulating the dynamic process of prostate tumor growth, invasion of nearby tissues, and distant metastasis in the lymph nodes, bone, and liver. The 4D mesh models (3D + time) are further incorporated into a realistic PET data generation chain to simulate realistic PSMA PET images over the specified time period.

Conclusions: Our initial results show the great potential of the model-based biophantoms in cancer diagnosis, staging, and treatment planning and response evaluation.