Author: Jiayi Du, Lihua Jin, Ke Sheng, Yu Zhou 👨🔬
Affiliation: Harvard University, University of California, San Francisco, UCLA, Department of Radiation Oncology, University of California, San Francisco 🌍
Purpose: Radiomics enables powerful insights into tumor biology through non-invasive imaging, excelling in diagnostic and prognostic predictions. However, due to a lack of mechanistic connections, questions about its biological foundation and reproducibility have prevented broader clinical applications. To enhance the understanding of image feature emergence and improve prediction processes, we developed a physics-driven computational model that simulates heterogeneous vascularized tumor growth and integrates Radiomics analysis.
Methods: We proposed a novel vascularized tumor growth model comprising a dynamic, functional, discrete vasculature system for nutrient supply and a continuum tissue component to simulate tumor growth and soft tissue deformation. Key biophysical parameters were calibrated using literature-reported data, allowing the system to evolve under physical constraints. The model generated functional maps of tumors, and quantitative features were extracted and analyzed to explore their biological significance.
Results: Large-scale in silico tumors, up to multi-millimeter diameters, were generated, replicating realistic vasculature morphologies and key functional and anatomical characteristics, including tissue perfusion, vascular density, oxygen levels, and tumor growth rates. Twenty tumor samples with random tumor proliferation rate (PR) and oxygen consumption rate (OCR) are generated, and correlation analyses revealed that the tumor PR plays a pivotal role in driving necrosis and tissue heterogeneity, while OCR dictates vascularization levels. We further demonstrated differences in the visibility of these biological processes on imaging features, identifying optimal imaging modalities for specific biophysical insights.
Conclusion: This work establishes a new paradigm by leveraging computational modeling to causally link underlying tumor biology with data-driven imaging features. It offers unique insights into tumor imaging strategies, enabling the selection of modalities tailored to specific biophysical parameters.