UCLA

Radiomics offers a promising method to discern tumor biology through non-invasive medical imaging, successfully performing various prediction tasks and demonstrating potential in clinical applications. However, limitations in interpretability and robustness are significant obstacles to its broad clinical adoption. In the era of personalized medicine, there is an urgent need to better understand the physio-biological properties reflected by Radiomics and, more fundamentally, the multiscale problem of how these microscopic tissue properties develop as the tumor grows, leading to macroscopic tumor patterns or features in medical images.In this study, we introduced a hybrid simulation platform that integrates continuum tissue dynamics with discrete vasculature modeling for large-scale, vascularized tumor growth simulations with comprehensive hemodynamic capabilities. Through innovative vasculature conditioning and remodeling strategies, this platform enables unbiased simulations of tumor development to sizes previously unattainable, closely mirroring the biophysical vascular properties and tissue growth patterns observed in actual tumors. Our study has significant implications for understanding tumor characteristics. By examining the influence of cellular proliferation rate (PR) and oxygen consumption rate (OCR) on tumor patterning and heterogeneity, we have elucidated the mechanistic links between biophysical properties and tumor characteristics. Key findings include the pivotal role of tumor proliferation rate in driving necrosis and tissue heterogeneity and the impact of OCR on tissue vascular density. Using our platform, we analyzed 20 randomly generated samples to predict PR and OCR using Radiomics, based on semantic and agnostic features. The resulting high-performance predictive model sees through tumor appearance to identify critical features that uncover underlying biological processes. Given the insight from modeling, the rationale behind feature selection can be understood, and features can be interpreted. Our study advanced our understanding of the complex tumor vasculature and tissue development problem and laid the groundwork for integrating computational models with Radiomics, bridging the gap between data-driven tumor prediction and fundamental biophysics. This integration opens up exciting new avenues for research in personalized medicine and beyond. It provides a new paradigm for interpreting tumor features and can help identify tumor property types with the highest potential to reveal specific biophysical properties, guiding the development and selection of imaging modalities for advanced non-invasive biophysical assessments.