Purpose: To elucidate the mechanistic relationship between the heterogeneous vascular tumor growth and its radiomics features, we developed an ultra-large-scale (ULS) hybrid mechanical tumor model that simulates biological and physical processes contributing to macroscopic tumor development.
Methods: In the ULS model, soft tissue was assumed mixed hyperelastic materials, combining a standard elasticity component and a growth component for volume changes due to cell proliferation and death. The tumor growth was supported by vasculature development driven by tissue-vasculature mechanical interaction and hypoxia-induced angiogenesis. A vasculature hemodynamics module determined the perfused region and provided wall shear stress for vessel remodeling. The growth of five baseline tumors and another five hypoxically adapted tumors were simulated. Metabolism intensity images with a voxel size of 503 μm were then synthesized for radiomics analysis.
Results: With identical initialization, at the end of the 12-day simulation, these two types of tumors significantly differ in volumes (16.9±0.3 mm3 vs. 20.9±1.5 mm3), necrosis volume faction (27.4±2.4% vs. 5.0±2.5%), mean oxygen partial pressure (15.8±0.7 mmHg vs. 20.6±3.2 mmHg), mean VEGF concentration (0.129±0.014 vs. 0.702±0.048), and mean vessel volume fraction (0.45±0.017 vs. 0.60±0.03). Statistically significant differences were observed in 434 out of 1130 radiomics features from the metabolic images of two tumor types. The top five features are ten-percentile gray level, GLCM maximum probability gray level, maximum grey level, GLSZM zone entropy, GLDM low gray-level emphasis, GLCM maximum probability gray level.
Conclusion: In this work, we developed a novel vascular tumor growth model capable of establishing a causal relationship between tumor heterogeneous growth patterns and texture features. The model reveals and explains significantly different image features between baseline and hypoxically adapted tumors. Therefore, our ULS model provides a mechanistic interpretation of radiomics features that have been used to empirically predict tumor characteristics and outcomes.
IM/TH- Foundational Skills: Feature extraction and texture analysis