Preliminary Computed Tomography Radiomics Model for Predicting Pretreatment CD8+ T-Cell Infiltration Status for Primary Head and Neck Squamous Cell Carcinoma.

Abstract

Immunotherapy has emerged as a treatment option for head and neck squamous cell carcinoma (HNSCC), with tumor response being linked to the CD8+ T-cell inflammation. The purpose of this study is to assess whether computed tomography (CT) radiomic analysis can predict CD8+ T-cell enrichment in HNSCC primary tumors. This retrospective study included 71 patients from a head and neck cancer genomics cohort with CD8+ T-cell enrichment status. Pretreatment contrast-enhanced neck CT scans were retrospectively reviewed using 3D Slicer for primary lesion segmentation.The SlicerRadiomics extension was used to extract 107 radiomic features. Ridge regression and lasso regression were applied for feature selection and model construction. Lasso regression defined Coarseness as the most important variable, followed by SmallDependenceEmphasis, SmallAreaLowGrayLevelEmphasis, Contrast.1, and Correlation.Ridge regression defined Coarseness as the most important variable, followed by SmallDependenceLowGrayLevelEmphasis, Contrast.1, DependenceNonUniformityNormalized, and Idmn. These variables identified by lasso and ridge regressions were used to create a combined logistic regression model. The area under the curve (AUC) for the lasso-generated model was 0.786 (95% confidence interval [CI], 0.532-1.000), and the AUC for the ridge-generated model was 0.786 (95% CI, 0.544-1.000). Combining the radiomic variables identified by lasso and ridge regressions with clinical characteristics including alcohol use, tobacco use, anatomic site, and initial T stage produced a model with an AUC of 0.898 (95% CI, 0.731-1.000). T-cell inflammation status of HNSCC primary tumors can be predicted using radiomic analysis of CT imaging and thereby help identify patients who would respond well to immunotherapy.