Topological and Texture-Based Feature Characterization of Post-Radiation Lymph Nodes in a Preclinical Model of Head and Neck Squamous Cell Carcinoma

Abstract

This study aims to characterize the immune architectural phenotype of cervical draining lymph nodes in preclinical HNSCC (head and neck squamous cell carcinoma) murine models using graph-based and texture-based pathomic features, and to evaluate their association with chemoradiation treatment outcomes.A cohort of 122 C57BL/6J mice was induced with one of three orthotopic tumor models: mouse oral cancer 1 (MOC1), mouse oral cancer 2 (MOC2), or murine lung metastatic 1 (MLM1). All mice were treated with cisplatin (5 mg/kg, intraperitoneal) and 8 Gy irradiation on days 0 and 7. Following exclusion of incomplete cases, a final cohort of 82 whole slide images was analyzed. Hematoxylin and eosin-stained lymph node specimens were digitized and processed in QuPath, where ten ROIs were placed per lymph node, each comprising 1% of the total lymph node area. A previously developed deep learning pipeline was used for automated lymphocyte detection and spatial graph construction, from which 18 topological features were extracted. Additionally, 54 texture features were extracted from the whole slide images using via second order matrices, including gray level co-occurrence matrix (GLCM), gray level run length matrix (GLRLM), and gray level size zone matrix (GLSZM). Features were evaluated via unpaired t-tests, LASSO regression, and multivariate Cox proportional hazards modeling to generate risk scores. LASSO regression selected 6 topological features (diameter, number of k-cores, average betweenness, ratio of betweenness, central node dominance, and cell density) which yielded significant survival stratification between high- and low-risk groups based on median risk score (p = 4.38e-6). For texture-based analysis, LASSO selected 8 features following correlation-based pruning, which similarly achieved significant risk stratification (p = 0.0022). High-risk stratification was associated with the MOC2 subtype and low-risk with MOC1, consistent with known radiosensitivity profiles of these models. These results demonstrate that graph-based and texture-based pathomic features of cervical draining lymph nodes are significantly associated with survival outcomes in irradiated HNSCC murine models, supporting lymph node immune architecture as a viable prognostic biomarker of chemoradiation response.