A Multiscale Radiopathomic Cox Modeling Framework for Survival Stratification In Murine Head and Neck Squamous Cell Carcinoma
Abstract
Purpose
To investigate mechanisms of chemoradiation resistance in murine head and neck squamous cell carcinoma using a multiscale framework integrating metabolic imaging (μPET), tissue heterogeneity (μCT), and cellular topology (digital pathology).
Methods
Oral cavity tumors were established in three syngeneic models with distinct immune contexts and treatment response: MOC1 (HPV-negative, indolent, CD8+ T cell-rich), MOC2 (HPV-negative, aggressive, CD8+ T cell-poor), and MLM1 (HPV-positive, mixed response). Tumor growth was monitored thrice weekly, and chemoradiation was initiated at ~50 mm³ with cisplatin (5 mg/kg) and image-guided radiation (8 Gy) on days 0 and 7. On day 14, metabolic activity was evaluated using 18F-FDG μPET (SUVmax normalized to liver), and μCT quantified tissue heterogeneity. Radiomic features capturing metabolic and structural variation were extracted. Post-necropsy, H&E slides were analyzed to derive pathomic features at local (cell–cell, 10.5 µm edge distance) and global (tissue architecture, 15.75 µm edge distance) scales. Associations among imaging features, topology, immune phenotype, and survival were assessed.
Results
A multiscale Cox model using 7 LASSO-selected features (1 CT radiomic, 4 PET radiomic, 2 pathomic) stratified high- and low-risk groups (log-rank p=0.0011). Pathomic analysis identified multiscale topological features associated with survival. At the cellular level, more tumor cell clusters at 10.5 µm indicated enhanced local connectivity. At the tissue level, increased cluster counts at 15.75 µm reflected altered higher-order organization. Architectural heterogeneity metrics—including cluster size variability, intercellular distances, k-core nodes, and correlation mean—showed mixed associations with outcome. CT radiomics features, e.g. size zone non-uniformity and long-run low gray-level emphasis, and PET features including GLCM contrast, joint entropy, GLRLM gray-level non-uniformity, and SUVmax, were linked to adverse survival.
Conclusion
Chemoradiation-resistant tumors exhibit coordinated disruptions across cellular topology, tissue heterogeneity, and metabolic texture. Multiscale imaging with radiopathomic analysis reveals prognostic utility, emphasizing integrated multiscale modeling for understanding tumor resistance mechanisms.