Identifying CT image radiomic biomarkers for predicting immunotherapy response of non-small cell lung cancer patients

Abstract

Checkpoint blockade immunotherapy (IO) provides improved long-term survival in a subset of advanced stage non-small cell lung cancer (NSCLC) patients. However, highly predictive biomarkers of IO response are an unmet clinical need. In this thesis, pre-treatment clinical covariates and quantitative image-based features (i.e., Radiomics) were utilized to identify parsimonious models that predict rapid disease progression (RDP) phenotypes and survival outcomes among NSCLC patients treated with IO. As part of the thesis, four studies were conducted. First, novel prognostic and predictive computed tomography (CT) radiomic features utilizing radial gradient and radial deviation maps were created. One feature, RD outside-border SD, was found to be associated with overall survival in two independent NSCLC cohorts. Second, clinical-radiomic models that predicted RDP phenotypes, including hyperprogressive disease (HPD), were created in the setting of NSCLC IO. Among 228 NSCLC patients, parsimonious clinical-radiomic models with modest to high ability (area under the curves: 0.812 and 0.843) to predict RDP were identified. In the third study, stable and reproducible peritumoral and intratumoral CT radiomic features of lung lesions were identified to reduce the chance of spurious findings. In the fourth and final study, pre-treatment clinical covariates and radiomics were utilized to identify a parsimonious risk-model based on survival outcomes among 332 NSCLC patients treated with IO. The most predictive radiomic feature (GLCM inverse difference) was found to be positively associated with CAIX expression, using a gene-expression and an immunohistochemistry dataset.|Keywords : Radiomics, Lung cancer, Immunotherapy.