A GPU-Based Computational Platform for Multi-Scale Modeling of Tumor Growth and Treatment Response In Lung Cancer Radiotherapy
Abstract
Purpose
Tumor growth and response to radiotherapy (RT) is a multiscale biological process encompassing cellular dynamics, extracellular transport, vascular and immune modeling. This work develops a comprehensive simulation platform to model this process with GPU acceleration.
Methods
We initialized from anatomically realistic lung geometry derived from a publicly available high-resolution human organ phantom. Pulmonary vessels were segmented to define the initial vascular configuration. A 5 cm cubic region of interest was extracted and simulated with four tightly coupled layers. Cellular layer modeled tumor cell density and phenotype transitions (proliferative, quiescent, apoptotic) on a voxel grid, driven by local oxygen and glucose availability. Extracellular oxygen, glucose, and VEGF fields were governed by diffusion–reaction equations with vessel-based source terms. Angiogenesis was simulated using an off-lattice vascular network with VEGF-guided, stochastic endothelial tip migration. Immune effects were represented by a transient tumor-associated antigen field induced by apoptosis. Radiation response was modeled using an oxygen-enhancement–ratio–modified linear–quadratic model. The entire framework was implemented on GPUs to enable efficient long-term (months) simulations over large spatial domains.
Results
Model parameters governing tumor growth were tuned to match experimentally measured tumor volume evolution from a syngeneic mouse lung tumor model. Comprehensive simulations were then performed over a 242-day period, including 180 days of tumor growth followed by fractionated RT (2 Gy per fraction, 30 fractions over 6 weeks). During the growth phase, tumor expansion was observed along with increasing hypoxia, elevated VEGF expression, angiogenic remodeling, and dynamic immune signaling. During and after RT, reductions in tumor size and corresponding changes in vascular structure, hypoxia, and VEGF distributions were observed.
Conclusion
We developed a GPU-accelerated, anatomically grounded, multi-layer simulation framework for modeling tumor growth and RT response. The model provides a powerful platform for studying tumor–microenvironment interactions and exploring treatment response mechanisms in silico.