Fig. 1

Heatmap of radiomic derived features (created using R programming language (ver. 3.5.1). Unsupervised hierarchical clustering identifies two distinct, and statistically significant (p value for phenotype split for two phenotypes, p = 0.032) tumor radiomic phenotypes in the group of patients with non-squamous carcinoma (n = 210). Chi square test p values quantifying the association of these phenotypes with BMI, smoking status and PDL1 expression are also included in the figure.